Classification (Machine Learning)

Traditionally, Enterprise Resource Planning (ERP) systems have relied on static rule-based mechanisms and batch analytics which limits their capacity to react to dynamic enterprise environments. This paper presents the design, implementation and experimental validation of an Adaptive AI-Driven Enterprise Resource Planning (A2ERP) framework that combines real-time data ingestion, online ensemble learning, explainable AI and scalable microservices architecture. This is not like traditional ERP enhancements that add stand-alone AI modules, but rather embeds adaptive intelligence right in the heart of the architecture. The implementation is tested with enterprise simulation datasets from finance, supply chain, HR and CRM modules. Experimental results show that its predictive accuracy (94.6%) is higher, its latency (95 ms) is lower, its throughput (1650 transactions/sec) is higher and its scalability is better than rule-based and batch ML ERP systems. The proposed framework establishes a practical pathway toward intelligent, transparent, and real-time strategic enterprise operations.

This book exists because we have watched too many AI initiatives fail for preventable reasons. Not because the models were poorly designed, not because the algorithms were flawed, but because the data that fed those models was fragmented, inconsistent, and ungoverned. We have watched organizations spend millions of dollars building machine learning systems on data foundations that could not support them, only to discover, after months or years of development, that the outputs could not be trusted because the inputs were never trustworthy. We have also watched organizations build excellent master data management programs that transformed their data quality and governance capabilities, enabling AI initiatives that delivered real value because they operated on a foundation of trusted data. The difference between these outcomes was not talent, budget, or technology. It was the presence or absence of a systematic, sustained commitment to data governance as the prerequisite for everything else. The intellectual foundation of this book is built on five internationally published research papers and three international patent applications that were filed in 2022. These publications represent years of research, experimentation, and production validation. The anomaly detection framework, the NLP metadata enrichment pipeline, the evidence-mapped governance architecture, the regulatory intelligence engineering approach, and the human-supervised AI control architecture described in these papers have all been tested in production environments where the cost of failure was measured in regulatory penalties, patient safety risks, and financial losses. We wrote this book for the technology leaders, data governance professionals, and MDM practitioners who are responsible for building the governed data infrastructure that enterprise AI requires. We wrote it to be comprehensive enough to serve as a reference for experienced practitioners and accessible enough to serve as an introduction for professionals who are new to MDM. We wrote it to bridge the gap between the academic literature on data governance and the practical realities of building and sustaining governance programs in complex organizational environments.
— 6 —
Most importantly, we wrote this book because the work it describes matters. In an era where AI-driven decisions affect credit access, healthcare outcomes, regulatory compliance, and customer experience, the quality and governance of the data that feeds those decisions is not a technical concern. It is an ethical obligation. We hope this book helps you fulfill it.

Background: Bladder cancer (BC) is a life-threatening malignancy that can be successfully treated if diagnosed in its early stages. Machine learning techniques, by using large biological databases, are suggested as important approaches for identifying accurate diagnostic biomarkers. The present study aimed to introduce a simple and accurate model for the diagnosis of BC.
Methods: RNA-sequencing information of 412 primary bladder tumors versus 19 normal bladder tissues from The Cancer Genome Atlas were analyzed using the TCGAbiolinks R package to identify differentially expressed genes (DEGs). Gene ontology properties and the corresponding pathways of DEGs were investigated using the online ShinyGO tools. To develop a diagnostic model for BC, two binary classifier machine learning algorithms, C5.0 and CHAID, were employed in three subgroups of train, test, and validation sets using SPSS Modeler version 18.1. Their efficacy was evaluated using performance measures for binary classification.
Results: Most of the identified DEGs were associated with microtubule organization, coagulation, and myelination. Based on the constructed models, four important RNAs (Tubulin Polymerization-Promoting Protein: ENSG00000171368, Proteolipid Protein-1: ENSG00000123560, RP11-473E2: ENSG00000228877, and Coagulation Factor X: ENSG00000126218) were identified as important classifiers for diagnosis in both C5.0 and CHAID models. The CHAID model demonstrated superior performance in the testing dataset, achieving an accuracy of 98.75%, an F1-score of 99.36%, and an AUC of 99.4%.
Conclusion: According to the results, machine learning algorithms are beneficial for the diagnosis of BC and potentially useful for improving personalized medicine in BC patients. The developed model may serve as a non-invasive, data-driven tool to support early diagnosis and personalized treatment planning in clinical settings. Further evaluation using laboratory tests is suggested to validate the obtained results.

The classification's traffic is regarded as a significant study domain because to the rising demand among network users. In addition to improving the identification of network services and addressing difficulties related to the security of traffic networks, it also simplifies and improves the accuracy of a broad variety of Internet application modes and activities. Over the course of the last several years, a multitude of traffic classification algorithms have been developed alongside their effective implementation. This paper proposed evaluate the classification issues of information security which happened through networks traffic by using machine learning (ML) classification algorithm the RamdomForst by using NSL-KDD test and NSL-KDD train dataset to show the performance of testing data and training data.

Hepatitis is a serious international health subject that may be prevented by way of early identity. However, conventional strategies for diagnosing hepatitis can once in a while be intrusive or inaccurate. Because Random Forest performs properly with complicated records and has built-in feature significance analysis, this painting suggests a hepatitis prediction framework based on its device mastering method. Comprehensive facts pretreatment changed into part of the technique, which included handling outliers, handling lacking values, and utilizing SMOTE to correct magnificence imbalance. The statistical Filter technique (SelectKBest) and versionexplanatory SHAP have been the two function choice strategies that were compared. The model that used SHAP as a function selection methodology beat the other strategies, as proven by experimental outcomes at the HCV-Egy-Data dataset. It accomplished up to 98% accuracy and greatly improved precision, do not forget, and F1-rating. In order to improve prediction overall performance in scientific diagnostics, this study highlights the significance of integrating sturdy preprocessing into interpretable machine learning.

Educational institutions generate large volumes of student-related data, including attendance records, internal assessment marks, assignment scores, and semester examination results. However, this data is often underutilized for proactive academic decision-making and early intervention strategies. Traditional approaches rely mainly on end-semester evaluations, which limit the ability of educators to identify students who are at risk of poor academic performance in a timely manner. This research aims to apply predictive analytics and machine learning techniques to forecast student academic performance and to identify at-risk students at an early stage of the academic cycle. The proposed system utilizes historical student data collected from academic records and applies data preprocessing techniques such as data cleaning, normalization, and feature selection to improve model performance. Classification algorithms including Decision Tree, Logistic Regression, and Naïve Bayes are implemented to build predictive models. The performance of these models is evaluated and compared using standard evaluation metrics such as accuracy, precision, recall, and F1-score. The experimental results indicate that machine learning-based predictive models can effectively classify students into performance categories and accurately identify those who are likely to face academic difficulties. The findings of this study highlight the practical value of predictive analytics in educational environments. By identifying at-risk students early, institutions can design targeted academic support mechanisms such as mentoring programs, remedial classes, and personalized learning plans. Furthermore, the proposed approach supports data-driven academic planning, improves student retention rates, and enhances overall learning outcomes. The research demonstrates that integrating predictive models into academic information systems can significantly contribute to improved student performance, institutional efficiency, and informed decisionmaking in higher education.

Aerosol categorization is vital for understanding atmospheric processes and improving satellite retrievals. Machine learning (ML) utilizing optical and microphysical properties enables efficient and automated aerosol classification, thereby enhancing air quality and climate modeling. This study presents an ML-based aerosol categorization scheme for Kanpur, utilizing AERONET (AErosol RObotic NETwork) data from 2001 to 2022. Thirteen key features, including Single Scattering Albedo (SSA), Particle Linear Depolarization Ratio (PLDR), Imaginary part of Refractive Index (Img-RI) at multiple wavelengths, and Extinction Angstrom Exponent (EAE), were selected based on Random Forest (RF) feature importance. The Synthetic Minority Over-sampling Technique (SMOTE) was applied to address class imbalance, and all features were standardized. Nine ML classifiers-Decision Tree (DT), RF, Gaussian Naive Bayes (GNB), K-Nearest Neighbors (KNN), Support Vector Classifier (SVC), AdaBoost, XGBoost, LightGBM, and CatBoost-were trained using 10-fold cross-validation and grid search. DT outperformed all models with the highest accuracy (0.997), precision (0.998), recall (0.996), F1-score (0.997), and perfect specificity (1.000), supported by high MCC (0.997), Kappa (0.997), AUC (0.998), balanced accuracy (0.996), and low log loss (0.094), indicating robust and reliable classification. Boosting models such as AdaBoost and RF also showed strong performance, while GNB and KNN underperformed. DT was further applied to nine additional AERONET sites across India for aerosol classification. The results demonstrate that tree-based and boosting models effectively classify aerosols, offering potential to enhance satellite aerosol retrievals. Seasonal analysis over Kanpur revealed that dust-driven aerosols peaked during the premonsoon season, while higher absorbing pollution aerosols were observed in winter, highlighting physically meaningful aerosol behavior linked to meteorology and emissions. This reinforces the relevance of the classification for air quality and climate applications.

This study examined the factors influencing the adoption of social media and consequent impact on agricultural distribution in the regions of mountainous districts Kullu and Shimla in Himachal Pradesh, India. Social media has come to light as a revolutionary tool that offers farmers the ability to predict the traditional barriers in the agricultural value chain. This research provides the critical insights into the drivers and barriers of social media adoption by the farmers, where the most influencing factors have been explored to investigate the level of adoption of social media and its impact on farmers and their income growth after using it for a purpose over the years. It employed the primary survey design and data were collected with the support of structured questionnaire using statistical techniques. Results indicate that the farmers adoption rate for social media in these regions has increased in the last three years and seek guidance to adopt it for more agricultural activities. Also, technical literacy, lack of guidance, lack of skills and poor internet access in some regions have been major problems. By using correlation in this study, relationships among number of years of experience using social media, frequency of using social media daily, education level of farmers have been some of the key variables. Technology Acceptance Model (TAM) has been used in this study to carry out internal statistically significant relationships between the hypotheses and to check behavioral intention (BI) of adoption of social media by the farmers as a marketing tool for their agricultural produce. Perceived usefulness (PU) and perceived ease of use (PEOU) demonstrated positive associations on behavior intention (BI) for the farmers of mountainous regions of Himachal Pradesh.

With the rapid growth of e-commerce and online shopping platforms, product reviews have become a crucial resource for consumers making informed purchasing decisions. These reviews contain valuable insights into customer sentiments and opinions about products and services, significantly influencing other customers' choices. However, manually analysing large volumes of reviews is time-consuming and challenging. Machine learning offers an automated solution by extracting and classifying sentiments expressed in reviews. The primary objective of this research was to develop a sentiment analysis model using supervised machine learning to classify sentiments as positive or negative. The Amazon Fine Food dataset from Kaggle was utilised, comprising 568,454 reviews, from which a random sample of 5,685 reviews was drawn for analysis. Data preprocessing involved removing special characters and HTML tags, handling missing values, and eliminating stop words. Five supervised machine learning models were employed and compared: Logistic Regression, Random Forest, Decision Tree, Naive Bayes, and Support Vector Machine (SVM). The dataset was partitioned into training (75%), validation (25%), and test sets (20% of the total dataset). Models were evaluated using accuracy, precision, recall, F1-score, and classification reports. Logistic Regression achieved the highest accuracy (64.47%) and demonstrated the best overall performance. Aspect-based sentiment analysis was also explored to provide granular insights into specific product features. Results were visualised using word clouds, pie charts, bar charts, and heat maps. Logistic Regression was selected for deployment, with the model and TF-IDF vectorizer saved using the pickle library for efficient processing of new data. This research assists both consumers and businesses by facilitating informed decisions and enhancing product quality. Future work could explore deep learning architectures, advanced feature engineering, and resampling techniques such as SMOTE and ADASYN to address class imbalance.

This study aims to investigate the impact of AI on reverse logistics in the construction sector, and focuses specifically on how sustainability performance plays a mediating role. The research is based on the RBV and DCT theories, to investigate how AIpowered processes are associated with sustainability practices; as well as reverse logistics optimization. This study collects primary data through structured questionnaires from key professionals in the construction industry. To evaluate AI, sustainability performance and reverse logistics concepts of this relation a Partial Least Squares Structural Equation Modeling (PLS-SEM) technique is used in the study. The findings reveal that AI is used to improve reverse logistics processes, while an improvement in sustainability performance is this process's most important intermediate consequence. The relationship between AI and sustainability magnifies the need to embed sustainable behaviors in AI-driven logistics strategies for better business productivity and environmental performance. The study emphasized that AI technologies must be adaptable to sustainability outcomes which can offer firms unique advantages in an increasingly complex and environmentally aware industry. This research contributes to the literature; this research seeks to contribute by providing empirical evidence concerning sustainability performance as a mediating variable explaining how AI influences reverse logistics. The content provided in this article will help industry practitioners and policymakers to benefit from AI for sustainable logistics management of construction practices.

The single most consequential question in modern utility-scale solar procurement is no longer which technology
has the highest nameplate efficiency, but which has the most predictable 25-year energy yield. Procurement teams
negotiating multi-hundred-megawatt contracts with project life of two and a half decades face a fundamental
information asymmetry: PERC technology has accumulated more than ten years of field service data across
multiple climate zones and millions of installed modules, while the leading candidate replacement technologies -
TOPCon and heterojunction (HJT) - have substantially less field history at scale. The reliability claim made by
either replacement technology rests largely on accelerated test outcomes and limited field deployments, neither
of which can substitute for the statistical confidence available from PERC field data alone.
This Field Engineering Report presents a structured technical comparison of bifacial PERC module reliability
against TOPCon and HJT alternatives, drawing on field measurement data from 14,260 PERC modules deployed
across five US climate zones over a service period extending to 10 years. The principal findings, summarized in
the bullet points below, support a clear engineering conclusion: bifacial PERC is currently the most fieldvalidated, climate-tolerant, and BOM-stable technology available at gigawatt-scale production, and replacement
by either TOPCon or HJT in projects with strict warranty exposure should be undertaken only with explicit
acknowledgment of the residual reliability uncertainty.

This thesis presents a comprehensive predictive maintenance system and application interface that integrates deep learning and blockchain technologies in order to enhance maintenance strategies in industrial systems. Traditional predictive maintenance systems have significant issues regarding data security and decentralization. This study aims to address these limitations by leveraging blockchain technology, with a specific focus on improving the reliability and verifiability of predictive maintenance processes.
In this study, an LSTM-CNN hybrid model was developed to evaluate complex patterns in both time and features, thereby enabling high-accuracy fault prediction. The proposed model is designed to perform binary classification for fault prediction in industrial equipment. During the implementation phase of the study, an open-source dataset was used to train and test the developed model. The Randomized Search method was used in the hyperparameter optimization process to increase the prediction success of the proposed model. The hybrid model was trained with 5-fold cross-validation, and class weighting and threshold value optimization methods were applied to eliminate the class imbalance problem. In the threshold optimization phase, F1-score-based methods are applied to maximize recall at three predefined minimum precision levels (0.05, 0.2, and 0.85), while identifying the most balanced trade-off between precision and recall. In the proposed system, sensor data are stored in a database (SQLite3), and cryptographic proofs generated using zero-knowledge techniques are transmitted to the Ethereum network. The Poseidon hash function is used to ensure data integrity, and the Groth16 protocol is used for Zk-Snark proof generation. This approach enables secure verification of data validity without publicly disclosing sensor data and simultaneously addresses scalability concerns. The system architecture is designed to include manager, operator, and engineer nodes, and all smart contracts are implemented using Solidity. In addition, a graphical user interface is developed using the Tkinter library in Python.

The experimental results demonstrate that the proposed LSTM–CNN hybrid model produces successful outcomes in terms of fault prediction performance. According to scenario where the decision threshold is optimized based on the F1-score, the model achieves an accuracy of 0.987, an AUC value of 0.979, and an F1-score of 0.794.
In future studies, the proposed system is planned to be implemented on the Ethereum mainnet instead of a test network, with a comprehensive evaluation of on-chain operational costs. However, instead of Zk-Snark proofs, which have a centralized structure, the use of Zk-Stark proofs, which are transparent and do not violate the principle of decentralization, is planned.

Clinical monitoring systems operate independently, but accuracy may suffer because intelligent input validation and alarm justification are not included. Following the architecture of a human-centric reinforcement learning framework, a clinical monitoring platform for hospital wards that uses deep neural networks as foundational components has been constructed. The platform can adapt to any clinical environment through real-time monitoring, context awareness of the physiological condition of patients, and online learning from actual mistakes. Hospital-acquired conditions remain major issues of quality in health-care delivery. These complications, at least in part, are related to the complexity of the clinical environment, which can be difficult to oversee. Autonomous systems can help close that supervisory gap, but existing clinical monitoring systems tend to operate independently. Although these systems are multivariable, their focus remains on detecting anomalies in the monitored signals and warning medical personnel in case something goes wrong. However, these alarms do not consider the operation of other clinical systems, which might justify an alarm without a real need-the so-called crying wolf. Reinforcement learning algorithms, at least in theory, are able to learn to achieve a goal by shaping their behaviour with the help of other intelligent agents operating in the same environment. To achieve this goal in the hospital ward environment, an architectural framework has been developed. The basic assumption is that patients in hospital wards should be in stable physiological conditions. The main task of the proposed autonomous clinical monitor is, therefore, to prevent deterioration of patients' conditions. Although all monitored signals are relevant, the mission is not to detect every single anomaly but to provide intelligent supervision of all hospital clinical monitoring, telecommunications and warning systems. Most importantly, the proposed system is autonomous: it is able to monitor the context information of the environment and, on the basis of its physiological state, to shape the error feedback, both confirming and invalidating alarms raised by other systems. The entire process is conducted in real time. The developed platform is capable of complementing the current clinical monitoring systems autonomously, learning on the basis of its own mistakes, and its effectiveness has been verified in a closed-loop pipeline by combining it with state-of-the-art clinical monitoring, wireless telecommunication and alarm systems.

Power distribution networks are increasingly susceptible to faults arising from aging infrastructure, integration of renewable energy sources, and growing grid complexity. Accurate, real-time fault classification is paramount for minimizing outage durations, reducing operational costs, and ensuring the stability of modern smart grids. This paper proposes a Quantum Machine Learning (QML) framework that integrates Variational Quantum Circuits (VQC), Quantum Support Vector Machines (QSVM), and a novel quantumenhanced Long Short-Term Memory (QLSTM) architecture for multi-class fault detection and energy systems optimization. The proposed method encodes three-phase current and voltage measurements into quantum states via amplitude encoding, extracts entanglement-based feature representations, and classifies faults-including single line-to-ground, double lineto-ground, line-to-line, and three-phase faults-with high accuracy. Furthermore, a quantum annealing module is embedded to solve the unit commitment optimization problem under stochastic renewable generation. Simulation results on the IEEE 13-bus and 33-bus test feeders demonstrate that the proposed QML framework achieves a classification accuracy of 98.7%, a precision of 97.9%, and a recall of 98.3%, outperforming state-of-the-art classical deep learning baselines by up to 4.2 percentage points while reducing computational complexity from O(n 3) to O(log n) for certain subroutines. The quantum optimization module converges to near-optimal unit commitment schedules ∼ 35% faster than classical mixedinteger programming solvers. These results highlight the transformative potential of QML as a next-generation tool for intelligent power system management.

Identifying environmentally friendly materials is essential for tackling worldwide sustainability issues, but
conventional methods frequently face drawbacks such as excessive expenses and sluggish advancement owing to
the extensive range of potential designs and intricate connections between properties. We propose a unified machine
learning and big data approach to expedite the development of sustainable materials by methodically integrating
computational, experimental, and data extracted from existing literature. The framework consists of four key
components: multi-source data acquisition, rigorous preprocessing, advanced model development, and robust
validation. Information is gathered from various sources, processed, and standardized to achieve uniformity, with
the creation of informative attributes including atomic traits and structural features. A variety of machine learning
algorithms, such as Random Forests, XGBoost, and Graph Neural Networks, are applied to model nonlinear
relationships and atomic-level interactions, with hyperparameters fine-tuned via grid search. The models are
evaluated using cross-validation and metrics such as RMSE, MAE, and 𝑅
2
to ensure reliability. Our method shows
considerable promise in lowering the time and expense linked to materials discovery by supporting high-throughput
screening and predictive modeling. Moreover, the inclusion of synthetic data generation increases dataset diversity,
which leads to better model generalizability. The proposed framework not only propels progress in the realm of
sustainable materials but also establishes an adaptable model for additional disciplines requiring data-centric
innovation. This work addresses the gap between computational predictions and experimental validation, advancing
the creation of sustainable technologies that have extensive environmental and industrial implications.

The work outlines a model using supervised machine learning for the classification of personalities prediction, or the ability to distinguish between both types of personalities, based on behavioral data containing seven characteristics, such as social event attendance, stage fright, and time spent alone. The 2,900 samples of data will be one-hot encoded for categorical variables and standardized for numerical variables. Logistic Regressive, The use of support vector machines (SVM), Random Forest (RF), XGBoost, a Naive Bayes model, and Choice Tree were from the classification models subsequently employed, and GridSearchCV was used to fine-tune the pipelines. Gradient-based models with learning rates between 0.01 and 0.2 and a batch size of 32 were used for training. The best-performing models were Logistic Regression and the tuned models were convergent to 92.93, while the default scores varied between 87.24 to 92.93. Metrics for assessment such as the F1-score (0.93) and ROC-AUC (0.93) attest to the reliability and durability of the suggested personality rating method.

Heart disease, also referred to as cardiovascular disease (CVD), is a broad category of illnesses that obstruct the coronary arteries, impairing the cardiovascular system's ability to operate normally. They are in charge of blood circulation, oxygen delivery, and the delivery of vital nutrients to various bodily areas. Heart disease claims the lives of over 17.9 million people worldwide each year, or about 31% of all deaths. The alarmingly high death toll from cardiovascular disease worldwide highlights the urgent need for thorough study and creative remedies. The current research uses various cutting-edge techniques in a multimodal manner to address the detection of cardiovascular disease in response to this pressing need. A variety of machine learning models, such as random forests, decision trees, Naïve Bayes, knearest neighbors (KNN), and neural networks, were employed in conjunction with a voting ensemble technique to attain an accuracy of 89.2%. Therefore, in order to lessen the severe effects of cardiovascular disease, this study offers an improved strategy. Its adoption offers better individual health results and more effective control of the societal and financial challenges associated with cardiovascular disease.

Sentiment analysis and opinion mining is an area that has experienced considerable growth over the last decade. This area of research attempts to determine the feelings, opinions, emotions, among other things, of people on something or someone. In particular, this article discusses the task of determining the polarity of reviews using an ordinal classification technique called Barycentric Coordinates for Ordinal Classification (BCOC). The aim of this analysis is to explore the viability of the application of BCOC on the field of sentiment analysis. This method is based on the hypothesis that the ordinal classes can be represented geometrically inside a convex polygon on the real plane by using barycentric coordinates. A set of experiments were conducted to evaluate the capability and performance of the proposed approach relative to a baseline, using accuracy as the general measure of performance. In general, the method is competitive with the baseline. For sentiment analysis with four classes, the results show improvements in the overall accuracy.

Epistemic uncertainty in neural networks is commonly modeled using two second-order paradigms: distribution-based representations, which rely on posterior parameter distributions, and set-based representations based on credal sets (convex sets of probability distributions). These frameworks are often regarded as fundamentally non-comparable due to differing semantics, assumptions, and evaluation practices, leaving their relative merits unclear. Empirical comparisons are further confounded by variations in the underlying predictive models. To clarify this issue, we present a controlled comparative study enabling principled, like-for-like evaluation of the two paradigms. Both representations are constructed from the same finite collection of predictive distributions generated by a shared neural network, isolating representational effects from
predictive accuracy. Our study evaluates each representation through the lens of 3 uncertainty measures across 8 benchmarks, including selective prediction and out-of-distribution detection, spanning 6 underlying predictive models and 10 independent runs per configuration. Our results show that meaningful comparison between these seemingly non-comparable frameworks is both feasible and informative, providing insights into how second-order representation choices impact
practical uncertainty-aware performance.

Credal predictors are epistemic-uncertainty-aware models that produce a convex set of probabilistic predictions. They provide a principled framework for quantifying predictive epistemic uncertainty (EU) and have been shown to improve model robustness across a range of settings. However, most state-of-the-art (SOTA) methods primarily define EU as disagreement induced by random training initializations, which mainly reflects sensitivity to optimization randomness rather than uncertainty from more substantive sources. In response, we formulate EU as disagreement between models trained under different degrees of relaxation of the i.i.d. assumption between the training and test distributions. Building on this idea, we propose CreDRO, which learns an ensemble of plausible models via distributionally robust optimization. As a result, CreDRO captures EU arising not only from training randomness but also from informative disagreement due to potential train-test distribution shifts. Empirically, CreDRO consistently outperforms SOTA credal approaches on downstream tasks, including outof-distribution detection on extensive benchmarks and selective classification in medical settings.

Effectively estimating the uncertainty attached to neural network predictions thus becomes essential to improve robustness, reliability, and trustworthiness. This paper provides an overview of various methodologies for representing, quantifying, and distinguishing two major types of uncertainties (namely, 'aleatoric' and 'epistemic' uncertainty) in neural networks. The review covers classical probabilistic techniques such as Bayesian neural networks and deep ensembles, methods from generalized probability that leverage uncertainty representations such as Dirichlet distributions, belief functions, random sets, probability intervals, and credal sets, among others. Additionally, interval-based approaches employing interval models are also examined. We discuss the strengths and limitations of various methodologies and identify promising research directions for potential future exploration.

The poultry industry in the United Kingdom is characterized by large-scale, highly integrated production systems that require the efficient coordination of resources to meet growing demand while maintaining sustainability, quality, and animal welfare standards. M-Park UK, a leading poultry processing company, operates in a complex environment where production efficiency is influenced by numerous interdependent
operational parameters, including feed supply, energy consumption, labor hours, processing time, and quality control measures.
This study aims to analyze the relationships between these key production variables and develop accurate predictive models for production output using advanced machine learning techniques. Descriptive statistical analysis and correlation assessment were initially employed to understand the data distribution and interdependencies among the operational parameters. Two machine learning approaches – Support Vector Regression (SVR) and Gaussian Process Regression (GPR) – were used to predict production outcomes. The models were evaluated using standard evaluation metrics including R², explained variance score, mean squared error, root mean squared error, and mean absolute error, which were applied to both training and test data. SVR demonstrated superior predictive accuracy and a strong ability to generalize beyond the training data, while GPR showed near-perfect performance on the training data and robust results on new data, with the added advantage of providing probability-based predictions. This study contributes to the growing body of research on data-informed decision-making tools for the agri-food sector and illustrates how machine learning techniques can be effectively implemented in industrial-scale poultry processing facilities.

The manual diagnosis of diabetic retinopathy (DR) is often invasive, time-consuming, expensive, and prone to human error. Additionally, it can be subjective, depending on the clinician's professional experience. Recently, automated computer-aided diagnosis (CAD) systems have significantly reduced the time and effort required for diagnosis while achieving superior performance compared to traditional methods. Researchers have extensively explored deep learning (DL) and convolutional neural networks (CNNs) for diagnosing DR from fundus images, yielding promising results and offering a viable alternative to conventional diagnostic approaches. In this study, a hybrid model named EffNet-SVM is proposed for the classification of DR and no DR cases using retinal fundus images. The model is trained and tested using the Asia Pacific Tele-Ophthalmology Society (APTOS) dataset, which includes both DR and no DR images. The EffNet-SVM utilizes EfficientNetV2-Small for feature extraction from input fundus images, and the extracted features are then classified using a support vector machine (SVM) with a radial basis function (RBF) kernel. The EffNet-SVM model outperformed eight state-of-the-art DL models from the literature, achieving the highest accuracy of 97.26%. Performance metrics validate that the proposed hybrid model can be effectively integrated into CAD systems for the automated analysis of fundus images. INDEX TERMS Diabetic retinopathy, deep features, EfficientNet, hybrid CNN, support vector machines, transfer learning.

In this paper, we describe a tool coined as artificial intelligence-based student learning evaluation tool (AISLE).
The main purpose of this tool is to improve the use of artificial intelligence techniques in evaluating a student’s
understanding of a particular topic of study using concept maps. Here, we calculate the probability distribution of the
concepts identified in the concept map developed by the studen

The technological landscape in the fields of Autonomous Surface Vessels (MASS), Unmanned Aerial Vehicles (UAVs), and robotics is undergoing a fundamental transformation characterized by the convergence of artificial intelligence, materials science, and cyber-physical systems. The evolution of modern engineering has reached a critical turning point where the boundaries between biological resilience and mechanical robustness are beginning to dissolve. At the heart of this transformation lies a fundamental challenge: How can we protect the next generation of autonomous assets, such as Maritime Autonomous Surface Ships (MASS), Unmanned Aerial Vehicles (UAVs), or deep-sea robotic systems, against increasingly complex and multi-domain threats? This study, entitled "Integrated System Architecture for AI-Powered Autonomous Platforms," stems from the need to move beyond the 20th-century "passive shell" philosophy. Traditional armor and structural design have long relied on static mass and monolithic materials, which are becoming increasingly inadequate in an era defined by high-speed kinetic threats, electronic warfare, and extreme fuel efficiency demands. We propose a revolutionary framework: "Armor Immunity System." The need for next-generation platforms exhibiting superior strength-to-weight ratios while maintaining high survivability levels has led to the emergence of the "Armor Immunity System" concept. This architecture goes beyond traditional passive shielding, creating a dynamic, forensic, and reconfigurable protective framework. Like a biological organism, an autonomous platform must be capable of sensing structural trauma, identifying the "DNA" of its components for forensic integrity, and adapting its defensive posture in real time. By combining the latest advancements in AI-powered material computing, high-entropy alloys (HEA), and Level 5 Digital Twin synchronization, this research presents a technical blueprint for a self-aware, self-defending architecture. This study provides a comprehensive technical analysis and synthesis of the components required to realize such a system, integrating AI-powered material discovery, high-accuracy digital twins, additive manufacturing of functionally graded materials, and advanced cloaking metasurfaces. It serves not only as a technical synthesis for maritime and aviation professionals but also as a strategic manifesto for the future of cyber-physical survivability. It bridges the gap between theoretical materials science and the operational realities of 21st-century autonomous warfare and logistics.

This research paper is devoted to optimization of the shell-and-tube heat exchangers (STHEs) with air injection using machine learning tools to estimate thermohydraulic performance. Comparative study of four machine learning models, Random Vector Functional Link (RVFL), Support Vector Machine (SVM), K-Nearest Neighbors (KNN) and Random Forest (RF), to assess their usefulness in predicting key performance indicators of temperature of outlets and pressure drops were analysed. As put forth in the study, data preprocess, feature selection, and hyperparameter tuning are evident in enhancing the precision of such models. The obtained results reveal that RVFL is preferable with respect to predictive capabilities as it produced high correlation coefficients in predicting outlet temperature and pressure drops. The work proves that machine learning can have significant capabilities to optimize the design of heat exchangers, with the advantage of using data to guide the pursuit of energy efficiency and thermal performance of an industrial process.

Globalization has reshaped higher education, expanded access and increased students' mobility across borders. Student mobility acts as a bridge, bringing together diverse backgrounds and experiences under one umbrella. Universities having student from diverse backgrounds thrives in creativity, innovation, and collaboration. For instance, Harvard enrols students from over 147 countries , and Oxford has nearly 40% of

At the present world, one of the main sources of the news is an online platform like different websites and social media i.e. Facebook, Twitter, Linkedin, Youtube, Instagram and so on. However, due to the lack of proper knowledge or deliberate activity of some cunning people, fake news is spreading more than ever. People in general, struggling to filter which news to trust and which one to discard. Even the sly people take advantage of the situation by spreading false news and misleading the people. Natural Language Processing, one of the major branch of Machine Learning, the wealth of research is remarkable. However, new challenges underpinning this development. Here in this work, Naive Bayes Classifier, a Bayesian approach of Machine Learning algorithm has applied to identify the fake news. We showed, besides the algorithms, how the wealth of corpora can assist to improve the performance. The dataset collected from an open-source, has been used to classify whether the news is auth...

What is the smallest multilayer perceptron able to compute arbitrary and random functions? Previous results show that a net with one hidden layer containing N -1 threshold units is capable of implementing an arbitrary dichotomy of N points. A construction is presented here for implementing an arbitrary dichotomy with one hidden layer containing IN/d1 units, for any set of N points in general position in d dimensions. This is in fact the smallest such net as dichotomies which cannot be implemented by any net with fewer units are described. Several constructions are presented of one-hidden-layer nets implementing arbitrary functions into the e-dimensional hypercube. One of these has only 14Nldllel [log2(Nld)jl units in its hidden layer. Arguments based on a function counting theorem of Cover establish that any net implementing arbitrary functions must have at least Nellog,(N) weights, so that no net with one hidden layer containing less than Ne/(d log*(N)) units will suffice. Simple counts also show that if the weights are only allowed to assume one of n, possible values, no net with fewer than Nellog,(n,) weights will suffice. Thus the gain coming from using real valued synapses appears to be only logarithmic. The circuit implementing functions into the e hypercube realizes such logarithmic gains. Since the counting arguments limit below only the number of weights, the possibility is suggested that, if suitable restrictions are imposed on the input vector set to avoid topological obstructions, two-hidden-layer nets with O(N) weights but only 0(X6) threshold units might suffice for arbitrary dichotomies. Interesting and potentially sufficient restrictions include (a) if the vectors are binary, i.e., lie on the d hypercube or (b) if they are randomly and uniformly selected from a bounded region. %I 1988 Academic press, hc.

The proliferation of misinformation across digital platforms has become a critical challenge in modern information ecosystems. This study presents the design, development, and evaluation of an automated fake news detection system using Natural Language Processing (NLP) and machine learning techniques. A supervised learning framework was adopted using a labeled dataset of real and fake news articles. Text preprocessing techniques including tokenization, stop-word removal, and lemmatization were applied to enhance data quality. Feature extraction was performed using Term Frequency-Inverse Document Frequency (TF-IDF), enabling effective numerical representation of textual data. A Logistic Regression classifier was implemented and evaluated using standard performance metrics. The model achieved an accuracy of 91.8%, precision of 90.6%, recall of 92.4%, and F1-score of 91.5%, demonstrating strong capability in distinguishing fake from real news. A webbased prototype was developed using Django to enable real-time classification. The findings confirm that combining TF-IDF with Logistic Regression provides a computationally efficient and interpretable approach to fake news detection. The study contributes to scalable misinformation mitigation strategies and highlights opportunities for integrating advanced NLP models in future systems.

Knowing the existence of coronary heart disease is very important to reduce the risk caused. Coronary heart disease is influenced by many factors, in diagnose requires complex analysis. Many proposed the application of a machine-learning algorithm to diagnose/predict coronary heart disease, but have not given perfect results (excellent). The machine learning algorithm is used to classify someone affected by coronary heart disease or not based on factors that have been determined input. The results of diagnosis/prediction are not perfect due to misclassification that is still large.to reduce misclassification, bagging techniques are proposed. The classification algorithm used in the study is Random Forest. Experimental results show that bagging techniques can reduce misclassified predictions of coronary heart disease.

Supervision of academic performance is very important to ensure that students can complete their education on time. There have been many proposed applications of machine learning algorithms to predict students’ academic performance. Prediction is done by analyzing a dataset of historical academic of the student’s grade. The dataset which analyzed has many variables (features), this can increase complexity and decrease model performance because maybe not all features are relevant. We propose to implement the forward selection algorithm to select features that can improve model performance. The result shows that the performance of predictive models of students academic scores can improve with the application of feature selection.

Parkinson is a disease that is caused by nerve cell damage in the brain and incurable. Knowing about Parkinson disease is very important so that medical action can be taken to prevent Parkinson’s getting worse. The dataset that uses to analysis for Parkinson disease using machine learning algorithms has many features. The dataset with many features can increase complexity, but not all features have a positive influence on the results of the analysis. Irrelevant features can reduce model performance. This research proposes to apply feature selection to choose features that have a positive effect so that the performance of the model does not decrease. The experiment results show that the application of feature selection can lead to better model performance.

Bu çalışmada, akademik metinlerin YZ ile yazılıp yazılmadığını tespit etmek için kullanılan YZ detektörlerinin raporlarının güvenilirliği araştırılmıştır. Etik ve hukuki açıdan tartışmaya açık olan bu durum günümüzde hem yazar hem de yayıncılar için önemli bir sorun hâline gelmiştir. Sorunun temel nedeni, akademik çalışmalarda YZ'nin kullanımıdır fakat sorunu büyüten ve derinleştiren, YZ'nin, çalışmanın hangi aşamasında ve hangi şekilde kullanıldığının belirsizliğidir. YZ detektörleri burada devreye girerek belirsizliği gidermeye çalışmaktadır fakat bu yöntemin işlevselliği ve güvenilirliği de bir başka tartışma konusudur. Tartışmanın odağı, YZ detektörlerinin yazar aleyhine abartılı oranlar vermesi ve/veya yanlış raporlama yapmasıdır. İki aşamadan oluşan çalışmada yöntem olarak içerik analizi bağlamında doküman incelemesi yapılmıştır. İlk önce, yazım süreçlerinin hiçbir aşamasında YZ teknolojisi kullanılmayan çalışma metnini, "insan" veya "YZ tarafından yazılmıştır" şeklinde raporlayan sekiz farklı YZ uygulamasından analiz etmesi istenmiştir. İkinci aşamada, aynı metin benzerlik oranı için farklı iki program tarafından taranarak intihal oranına yönelik rapor elde edilmiştir. Analiz sonucuna göre, YZ detektörleri metnin büyük oranda YZ tarafından yazıldığını, benzerlik taraması yapan programlar ise intihal oranının oldukça düşük olduğunu raporlamıştır. Bu sonuçlardan YZ detektörlerinin yanıldığı ve büyük oranda hatalı raporlamalar yaptığı anlaşılmaktadır. Ortaya çıkan durum, akademik üretim sürecindeki tüm paydaşların farkındalığını artırarak daha ilkeli davranması gerektiğini göstermektedir. Çalışmanın özgün katkısı, söz konusu süreçlerin etik ve adalete daha uygun şekilde yürütülmesi için özne olan yazar ve bir araç olması gereken YZ ve sürecin diğer paydaşlarının konumlarının doğru biçimde netleştirilmesinin gerekliliğine dikkat çekmesidir. Sorunu aşmak için, farklı alanlarda benzer çalışmalar yapılmalı ve yayım süreci paydaşlarını ortak noktada buluşturacak politikalar geliştirilmelidir.///This study, designed as an experiment, concerns the detection of whether academic texts are generated using artificial intelligence (AI). This situation, which is open to debate from an ethical and legal perspective, has become a significant problem for both authors and publishers today. The root cause of the problem is the use of AI in academic work, but what exacerbates and deepens the issue is the uncertainty surrounding at what stage and in what manner AI is used. AI detectors come into play here, attempting to resolve this uncertainty, but the functionality and reliability of this method is another topic of debate. The focus of the debate is that AI detectors give exaggerated rates against the author and/or produce false reports. In this two-stage study, document analysis was performed as a method in the context of content analysis. First, eight different AI applications (AI detectors) were asked to analyze the study text, which did not use AI technology at any stage of the writing process, and report whether it was written by a “human” or “AI.” In the second stage, the same text similarity ratio was scanned by two different programs to obtain a report on the plagiarism rate. According to the analysis results, despite different variables, AI detectors reported that the text was largely written by AI, while similarity scanning programs reported that the text's plagiarism rate was quite low. When this seemingly contradictory situation is evaluated, it is understood that AI detectors are mistaken in this regard and produce largely inaccurate reports. This situation demonstrates the need for all stakeholders in the academic production process to increase their awareness and behave more ethically. The original contribution of this study is that it draws attention to the need to clarify the positions of the perpetrator, the author, and AI, which should be a tool, in order to ensure that these processes are conducted in a more ethical and fair manner. To overcome this problem, similar studies should be conducted in different fields, and policies should be developed that bring the stakeholders in the field to a reasonable middle ground.

The use of 3D printing for creating three-dimensional objects is rapidly advancing in the manufacturing sector. Artificial intelligence is playing a key role in enhancing the efficiency and quality of 3D printing processes. This article delves into the ways in which artificial intelligence is impacting 3D printing. Prior to commencing a printing job, a machine learning-powered printability checker is used to assess the feasibility of a 3D object. Parallel slicing algorithms are employed to speed up the slicing process, while intelligent path planning further optimizes the process. Intelligent algorithms are also used to match demand with resources, supporting a Cloud service platform and evaluation model that allow customers to access services on demand. Additionally, three machine learning algorithms are discussed, which are designed to identify defects in products during cyber-attacks. The article also highlights the potential for further research in areas such as multi-indicator printability, simplifying complexity thresholds, expediting prefabrication, real-time control, enhancing security, and detecting defects in custom designs, particularly within the industrial context.4.0.

With the rapid advancement in THz technology and continuously increasing number of electrical and electronic devices in this frequency band, highly efficient and functional THz shielding materials are required to ensure the intended working of these devices. Polymeric composites of hybrid nano-fillers including CNFs and magnetic nanoparticles possess distinctive capability of providing high shielding efficiency primarily absorptive in nature with field functional properties. In this article, lightweight polymeric nanocomposite coatings of continuous carbon nanofibers and magnetite (Fe 3 O 4), as potential shielding material for the THz frequency band, have been prepared using a cost-effective conventional air-spray coating method. The obtained results are quite exclusive, as the sample with 40 wt % Fe 3 O 4 outperforms at a very low thickness value of 540 μm, showing the highest SE value of ∼60 dB. These results are superior compared to the results obtained from most of the contemporary polymer composite materials, which usually provide good shielding at high thickness values in the mm range. Also, this study establishes that shielding performance of the polymeric hybrid nanocomposites (Fe 3 O 4 + CNFs) is substantially better than a single nanocomposite CNF filler previously synthesized and characterized through the same approach.

The increasing reliance on data-driven decision-making in modern enterprises has elevated the
importance of effective Master Data Management (MDM) as a foundation for trusted and intelligent business
operations. This research reimagines enterprise MDM by proposing a comprehensive framework that integrates
data governance, advanced analytics, and intelligent automation to enhance data quality, consistency, and
reliability across organizational systems. The study addresses critical challenges such as data silos, inconsistency,
and lack of trust in enterprise data by introducing a unified architecture that ensures standardized data definitions,
real-time synchronization, and secure data access. The proposed model incorporates policy-driven governance,
metadata management, and role-based access control to maintain data integrity and compliance with regulatory
standards. Furthermore, the integration of intelligent techniques, including automation and data validation
mechanisms, improves decision-making capabilities and operational efficiency. Experimental evaluation
demonstrates that the reimagined MDM framework significantly enhances data accuracy, reduces redundancy,
and supports scalable enterprise operations. The findings highlight the strategic role of advanced Master Data
Management in enabling organizations to build trust in their data ecosystems while driving innovation and
intelligent business transformation.

This methods note introduces DOL-Diff v0.1, a paired-output procedure for comparing ordinary model-assisted responses with responses generated under an explicit Directional Operating Layer. Rather than treating a single output as the final object of evaluation, DOL-Diff examines the difference between a baseline response and a DOL-active response produced from the same input. The DOL-active condition applies Anti-Goodhart constraints, anti-sterilization checks, Claim-Verb Guard, and strongest-defensible-bounded-signal principles. The resulting difference is interpreted as a differential operating trace. The aim is not to claim internal model change, autonomy, or validated mechanism, but to test whether explicit operating constraints can improve claim discipline, preserve bounded signal, and increase decision clarity in model-assisted research synthesis. This release provides a procedural framework, scoring rubric, templates, and supplementary materials for future blinded or semi-external evaluation.

Enterprise data centers are tasked with managing large-scale applications and data for multiple businesses, making them private distributed systems in the cloud computing paradigm. Data center operations are inherently risk-prone, and commissions and omissions often expose the data center and hosting customers to record fines for non-compliance with sector-specific security, privacy, and operational continuity standards. Failing to detect and fix issues in time can also lead to operational infractions that directly affect business continuity, security breaches, data leaks, and a multitude of other service disruptions. Therefore, risk detection and compliance remediation are missions conducted by specialized teams, following formal processes and often using dedicated tools. Although risk detection and compliance remediation are traditionally viewed as human-centric, autonomous solutions exist and bring associated advantages. Nevertheless, they rarely address standalone data center needs, either operating as unsupported external tools (e.g., vulnerability scanners) or targeting only some types of issues (e.g., anomaly detection, fault diagnosis, etc.). A new class of general-purpose, self-driving data center risk detectors conforms to a sensing-perception-reasoning-design-execution action and feedback loop and, endowed with proper control mechanisms, can automatically contain, remediate, and restore. Such detectors generally require either ad-hoc policy engines for compliance mandates or proactive prevention workflows. In agentic AI frameworks, supervisory activities such as auditing, verifying, and explaining detected incidents also benefit from automation.

The rapid increase in waste production has become a major environmental issue, particularly in urban areas where improper waste management leads to pollution and inefficiency. This study aims to analyze and compare the performance of K-Nearest Neighbors (KNN), K-Medoids, and Support Vector Machine (SVM) algorithms in waste classification within Smart Waste Management systems. The methodology involves data preprocessing, feature extraction, model training, and evaluation using accuracy metrics. Based on recent studies [2], [5], [6], SVM demonstrates superior performance in handling complex datasets compared to KNN and K-Medoids. Therefore, SVM is considered the most suitable algorithm for intelligent waste classification systems. Furthermore, the findings of this study highlight the importance of selecting appropriate machine learning algorithms to enhance the efficiency and accuracy of automated waste sorting processes. The implementation of SVM not only improves classification performance but also supports the development of sustainable and intelligent waste management systems. This research contributes to the growing field of Al-based environmental solutions and provides a reference for future studies aiming to optimize smart waste technologies through advanced data-driven approaches.

Clinical order normalization and semantic mapping are critical processes in emergency care, where rapid decision-making relies on accurate interpretation of heterogeneous and often unstructured clinical data. Traditional rule-based and statistical approaches have struggled to handle the variability, ambiguity, and contextual complexity inherent in electronic health records (EHRs). Recent advances in transformer-based models have demonstrated significant improvements in natural language understanding tasks, making them highly suitable for clinical text processing and concept normalization. Transformer architectures, particularly those leveraging self-attention mechanisms, enable contextualized representation learning that captures semantic relationships across diverse clinical terminologies and documentation styles. This study explores the application of transformer-based models for clinical order normalization and semantic mapping in emergency care settings. By integrating sentence transformers and domain-specific pretrained models, these systems can effectively map free-text clinical orders to standardized vocabularies such as the Observational Medical Outcomes Partnership (OMOP) common data model and LOINC ontologies. Furthermore, transformer-based approaches have shown strong performance in semantic similarity measurement, concept classification, and multimodal data integration, all of which are essential for improving interoperability and clinical decision support in high-pressure emergency environments. Overall, transformer-based models represent a transformative approach to clinical order normalization and semantic mapping, offering substantial potential to improve data standardization, interoperability, and decisionmaking efficiency in emergency care systems.

Accurate land use and land cover (LULC) classification remains challenging when visually and spectrally similar classes coexist, particularly in RGB-based remote sensing imagery. While numerous machine learning (ML) and deep learning (DL) models report high overall accuracy, limited attention has been given to the spatial structure and persistence of misclassification errors. In order to gain more insight into model-dependent and systematic error patterns, this study introduces a misclassification-driven evaluation perspective through a controlled cross-architecture comparison of custom Convolutional Neural Networks (CNNs), AlexNet, and ResNet50 models applied to the EuroSAT dataset under a fixed RGB-only setting. This approach reduces input-related variation and allows a controlled comparison of model behavior across architectures. We demonstrate that models with similar overall accuracy can exhibit different class-wise error distributions. Across the evaluated architectures, misclassification patterns were stable for specific confusable class pairs (notably Annual Crop vs Permanent Crop), indicating that RGB-only inputs impose a dominant limitation that persists across model families. These results show that improving the reliability of land-use and land-cover classification requires more than changing models; it also requires richer feature representation and greater data diversity. The paper identifies several shortcomings that must guide the next generation of multispectral and domain-adaptive research and provides practical suggestions regarding the improvement of accuracy and robustness of LULC classification systems.

Nowadays waste generated by every household is increasing drastically. So, we need a smarter way for collecting and managing the waste in an efficient manner. This model proposes an AIoT (Artificial Intelligence of Things) aided smart waste management system that leverages various sensors to monitor the level and weight of waste within bins. Upon reaching a predefined threshold, the system triggers an alert to relevant authorities, facilitating timely intervention and efficient waste management strategies. The primary objective of this project is to develop a sophisticated trash bin integrated with sensors and a mobile application. The application serves as a comprehensive platform for waste collection management, enabling users to track collection schedules, automatically segregate plastic waste, and even facilitate payment for garbage disposal services. Each bin is assigned a unique identifier for authorities to monitor its location and take necessary actions. By seamlessly transmitting sensor data to the designated app, this project holds the potential to support initiatives such as Haritha Karma Sena, an endeavor spearheaded by Kudumbasree, thereby streamlining their operations and contributing to a more sustainable waste management ecosystem.

Cardiovascular disease (CVD) remains one of the leading causes of death globally, underscoring the need for effective early risk prediction. This systematic literature review analyzes research published between 2013 and 2023 on the application of machine learning (ML) in CVD risk prediction. Key areas examined include feature selection, data preprocessing, algorithm choice, and model evaluation. Studies were selected from ACM Digital Library, IEEE Xplore, ScienceDirect, and Scopus based on predefined research questions. Common challenges include limited or low-quality datasets, inconsistent preprocessing methods, and the need for clinically interpretable models. Widely used algorithms include random forest (RF), support vector machine (SVM), decision tree (DT), logistic regression (LR), naïve Bayes (NB), k-nearest neighbor (K-NN), and extreme gradient boosting (XGBoost). The review highlights that robust preprocessing, optimal feature selection, and thorough model validation significantly improve predictive accuracy. It also emphasizes the importance of balancing performance with interpretability for clinical adoption. Finally, the study proposes a structured framework to guide future research and practical implementation, including the integration of genetic and behavioral data to support more personalized and effective cardiovascular care.

Penelitian ini bertujuan untuk melakukan klasifikasi penyakit serangan jantung menggunakan metode machine learning, yaitu K-Nearest Neighbor (KNN), Support Vector Machine (SVM), dan Decision Tree. Dataset yang digunakan berasal dari Kaggle yang berisi data pasien dengan berbagai atribut kesehatan seperti usia, jenis kelamin, tekanan darah, dan kadar gula darah. Sebelum proses klasifikasi dilakukan, data melalui tahap preprocessing berupa data cleaning dan normalisasi, kemudian dibagi menjadi data latih dan data uji dengan perbandingan 80:20. Hasil penelitian menunjukkan bahwa ketiga algoritma memiliki performa yang berbeda dalam mengklasifikasikan data. Decision Tree dan Support Vector Machine (SVM) cenderung memberikan hasil yang lebih stabil dengan tingkat akurasi yang tinggi, sedangkan K-Nearest Neighbor (KNN) memiliki performa yang cukup baik namun sangat bergantung pada pemilihan parameter. Secara keseluruhan, penelitian ini membuktikan bahwa metode machine learning dapat dimanfaatkan untuk membantu proses klasifikasi penyakit jantung secara lebih cepat dan akurat.

The paper proposes an empirical test of the structure of autopoietic cycles, previously developed in the author's theoretical works, using data from non-invasive recording methodsmagnetoencephalography (MEG) and scalp electroencephalography (EEG). The structure includes the following elements: a stable internal model D, its changes ∇D, a self-reference threshold C′, and a perception operator κin. Based on open MEG data (ds007353, n=30; COGITATE, subject CA124) and EEG data (ds002718, n=18; ds005429, n=15), operational correspondences of these elements adapted for non-invasive methods are proposed: κin is associated with the correlation between early (0.3-0.5 s) and sustained (0.5-0.7 s) gamma activity (30-50 Hz) in the prefrontal and posterior cortex, respectively; D is associated with stable categorical clustering in the posterior cortex (RSA); C′ is associated with ROC analysis of the threshold for discriminating target and standard stimuli in an oddball paradigm. The results confirm the predictions derived from the structure of autopoietic cycles. The operator κ_in exhibits an invariant negative sign: in the MEG dataset ds007353, all 30 subjects (100%) showed a negative correlation (mean-0.082 ± 0.038); in the COGITATE MEG data (subject CA124), the value of κin reached-0.584; in the EEG data, all 33 subjects (100%) from two independent samples (means-0.402 ± 0.067 and-0.413 ± 0.070, respectively) also showed a negative sign. The stable internal model D, estimated from the EEG dataset ds002718, reveals significant categorical clustering in the posterior cortex (mean RDM = 0.486 ± 0.058), with the largest distance between the categories "famous faces" and "scrambled" (0.806 ± 0.220). The self-reference threshold C′ in the EEG dataset ds005429 is close to zero (C′ ≈ 0), AUC = 0.505 ± 0.017, reproducing the results previously obtained with invasive recordings. Additionally, wavelet analysis was used to estimate gamma coherence between the prefrontal and posterior cortex (0.552 ± 0.072), which differs significantly between "famous faces" and "scrambled" (p = 0.0006) and correlates weakly with κin (r =-0.417), indicating the independence of these two characteristics. The obtained results demonstrate that the key elements of the structure of autopoietic cycles can be related to measurable empirical referents in non-invasive MEG and EEG data, opening up prospects for applying the proposed structure in consciousness research and cognitive studies. Detailed pseudocodes of the calculation algorithms are given in Appendix A.

Machine Learning merupakan bagian penting dari Artificial Intelligence yang memungkinkan sistem komputer untuk belajar dari data dan meningkatkan kinerjanya tanpa pemrograman eksplisit. Makalah ini bertujuan untuk membahas konsep dasar machine learning dengan fokus pada dua pendekatan utama, yaitu supervised learning dan unsupervised learning. Supervised learning menggunakan data berlabel untuk keperluan prediksi dan klasifikasi, sedangkan unsupervised learning digunakan untuk menemukan pola dan struktur dalam data tanpa label.
Pembahasan dalam makalah ini mencakup prinsip dasar, karakteristik, metode, serta perbedaan antara kedua pendekatan tersebut. Selain itu, dijelaskan pula algoritma yang umum digunakan seperti klasifikasi, regresi, clustering, association, dan dimensionality reduction. Hasil pembahasan menunjukkan bahwa supervised learning lebih efektif digunakan untuk prediksi berbasis data berlabel, sementara unsupervised learning lebih sesuai untuk eksplorasi pola dalam data.
Dengan demikian, kedua metode memiliki peran penting dalam pengambilan keputusan berbasis data dan telah banyak diterapkan dalam berbagai bidang seperti industri, bisnis, dan teknologi. Pemahaman terhadap kedua pendekatan ini menjadi hal yang penting dalam pengembangan sistem yang cerdas dan efisien.

This paper audits the claim that, under repeated task-transform conditions, a fixed familiar-task response schema is replaced by a stable representation-invariant rule mapping. The audit is conducted under a restricted learning scope: recurring problem classes presented under stable transformation families, fixed evaluation criteria, and explicitly declared pre-and post-mutation rule objects. Rather than presupposing transfer, understanding, or deeper learning, the paper forces the declaration of recurrence and structure and evaluates whether both are admissible under stable reference conditions. It then applies Rule-State Separation and a proxy discipline that prohibits structural inference from grades, speed, familiarity, or local correctness alone. Under the declared conditions, the relevant rule object is explicitly modified: a formatbound response schema is replaced by a stable rule mapping that reconstructs across symbolic, verbal, graphical, or novel variants of the same task class. The resulting structural classification is therefore: Ψ ≠ 0. A local transfer-rule mutation claim is licensed under the declared conditions.

Advanced Structural Health Monitoring (SHM) and digital performance evaluation have emerged as critical strategies for enhancing the resilience and service life of civil infrastructure. This study investigated the integration of Digital Twins (DT), Internet of Things (IoT) sensors, and artificial intelligence (AI) algorithms to monitor, assess, and 467 Global Research journal of Natural Science and Technology (Grjnst) Volume 3 Issue 1 (2025) predict structural performance in real time. Traditional inspection methods were found to be episodic, subjective, and limited in detecting early-stage structural anomalies. The research employed a combination of IoT-enabled sensor networks, multi-modal data integration, and AI-based predictive modeling to identify deterioration patterns, simulate dynamic load responses, and optimize maintenance planning. Results demonstrated that DT-enabled systems significantly improved condition assessment accuracy, facilitated proactive maintenance, and reduced operational risks associated with infrastructure failure. Machine learning models accurately predicted structural degradation trends, while multi-modal integration of remote sensing, satellite, and ground-based data enhanced the contextual understanding of complex infrastructure systems. The study also identified key challenges, including high implementation costs, data interoperability issues, and cybersecurity concerns, which may hinder widespread adoption. Recommendations focused on standardized deployment protocols, long-term monitoring strategies, and integration of emerging technologies to support scalable and cost-effective SHM systems. Overall, the findings underscored the importance of data-driven, predictive frameworks in transforming infrastructure management from reactive to proactive approaches, enabling optimized service life, improved resilience, and informed decision-making.