A Comparative Analysis of Classification Algorithms for Extrovert and Introvert Identification

International Conference on Innovative Practices in Technology and Management (ICIPTM 2026) 2026 5th International Conference on Innovative Practices in Technology and Management (ICIPTM) | 979-8-3195-4328-8/26/$31.00 ©2026 IEEE | DOI: 10.1109/ICIPTM69057.2026.11465760 A Comparative Analysis of Classification Algorithms for Extrovert and Introvert Identification Prashant Agrawal Department of Computer Applications Krishna Institute of Engineering & Technology (KIET) Ghaziabad, Delhi-NCR, UP, India [email protected] Divas Tewari Graphic Era Hill University Bhimtal; Centre for Promotion of Research, Graphic Era (Deemed to be) University Dehradun, Uttarakhand–248002, India [email protected] Shokir Ataev Department of Law Urgench State University Urgench, Uzbekistan [email protected] Orcid: 0009-0007-4658-2631 Charosxon Sabirova Department of Pedagogy and psychology Urganch Innovatsion university Urgench, Uzbekistan [email protected] Orcid: 0009-0007-0193-7603 Abstract—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. Keywords—Behavior prediction, grouping of introverts as well as neural networks, SVM, xg boost, gridsearchCV, design of features, one-hot encoding, standard scaler, ROC-AUC, accuracy, the F1- score behavioral data analysis, and personality. I. INTRODUCTION Since personality prediction is used in hiring, healthcare, education, and personalized recommendations, it has been given top priority in the fields of psychology, AI, and data science. Because they have a direct impact on how people communicate, make decisions, and interact with others, quiet and extrovert personality types have attracted a lot of attention among personality traits [1]. Conventional methods use psychometric tests and questionnaires, which are laborious and frequently arbitrary [2]. Using behavioral, textual, and multimodal data, personality can now be automatically predicted thanks to recent developments in data mining (ML) together with deep learning (DL) [3]. It seems that there are plenty of models such as LR, a SVM, RF, Naive Bayes and XGBoost in particular that have been demonstrated to perform well in classification problems [4]. Model generalization and accuracy are also improved by hyperparameter tuning methods such as GridSearchCV [5] [6]. streamlined models have been validated to have an accuracy of above 90 per cent Barno Matchanova Department of National Idea and Philosophy Urgench State Pedagogical Institute Urgench, Uzbekistan [email protected] ORCID: 0009-0004-5217-6435 Sarvarbek Matniyazov Department of History Mamun University Khiva, Uzbekistan [email protected] while identifying personality traits [7] [8]. The study presents the best-performing model by comparing this classification with both untuned and tuned machine learning models [9], [10]. II. LITERATURE REVIEW The natural language processing highlighted effectiveness of convolutional neural networks (CNNs) for emotion detection in textual data. In [11], a hybrid CNN-LSTM model was proposed to improve contextual emotion detection, demonstrating that combining CNN with LSTM enhances classification accuracy (92.4%) and F1-score (91.7%). Similarly, [12] showed that CNNs alone are effective in handling short text sentiment and emotion detection tasks, achieving an accuracy of 90.2% and precision of 89.6%, thereby proving CNN’s efficiency in capturing n-gram level features. Elaborating on this, [13] showed that CNNs that have pre-trained GloVe embeddings can be used to classify the emotions expressed in the text without manual feature engineering and still have high accuracy and recall rates of 91.8% and 92.1% respectively. Additionally, [14] proposed a multi-channel CNN model to learn fine-grained emotional indicators using texts in social media, and it achieved higher accuracy (93.5%) and F1-score (92.9) with CNN being able to adapt to harsh, user-generated information. A more recent paper [15] introduced a CNN architecture including a semantic embedding layer, obtaining 94.1 and 93.8 percent accuracy and precision respectively, which showed that deeper semantic representation enhanced performance in finegrained classification tasks. All these studies confirm the strength of CNN in detecting text-based emotions. TABLE I. LITERATURE SURVEY SUMMARY Ref. Key Learnings Techniques Employed Performance Metrics with value [16] Ensemble methods improve robustness in personality classification tasks Random Forest, XGBoost Accuracy = 92%, F1 = 0.91 979-8-3195-4328-8/26/$31.00 ©2026 IEEE Authorized licensed use limited to: Chandigarh University. Downloaded on April 21,2026 at 09:24:50 UTC from IEEE Xplore. Restrictions apply. [17] SVM with optimized kernels yields superior classification in small datasets SVM (linear, RBF kernel) Accuracy = 93%, ROCAUC = 0.94 [18] Text-based embeddings enhance introvert– extrovert trait detection Logistic Regression + TF-IDF Accuracy = 91%, Precision = 0.90 [19] Deep learning models capture contextual dependencies in personality text data LSTM, BiLSTM Accuracy = 94%, F1 = 0.92 [20] , [21] Feature engineering and hyperparameter tuning improve classical ML models GridSearchCV with Decision Tree, Random Forest Accuracy = 90%, Recall = 0.89 III. DATASET The statistics employed in this study consist of the data on personality traits taken to classify people into Introverts and Extroverts in a binary system. The data set has 2,900 records which have various behavioral and demographic attributes that lead to prediction of personality. Of these, there are about 1,520 cases that are termed as Introverts and a corresponding 1,380 cases that are termed as Extroverts, which is relatively even as to the two classes. The characteristics are a mix of numbers like age, activity levels, and frequency of interaction, and categories, which are gender and occupation, as they allow flexible-type of learning setting. To pre-process the data, missing data were handled with the help of the default function, categorical data were coded with the help of the OneHot Encoding, and numerical features were normalized with the help of the StandardScaler in order to normalize the distributions. To maintain generalization, the data was separated into 80% training (2,320 shots) and 20% testing (580 shots). In addition, random sampler of 300 shots was also used in visualization exercises like pair plots. This data set offers a solid basis on which various machine learning models can be estimated. IV. Fig. 1. Proposed Methodology flowchart. TABLE II. Layer Parameters/Units Details Input Layer Input StandardScaler (numeric), OneHotEncoder (categorical) Inputs Normalizes numerical values; Encodes categorical features into binary format Train-Test Split 80% Train, 20% Test Ensures model evaluation on unseen data Logistic Regression Solvers=[liblinear, lbfgs] SVM Kernel, Gamma Random Forest Estimators XGBoost Estimators, LR, Subsample Naïve Bayes Var Smoothing Decision Tree Evaluation Layer Max Depth, Min Split Regularization strength and solver optimization Controls margin, kernel type, and gamma scaling Controls tree depth, splits, and ensemble size Gradient boosting with depth, learning rate, and sampling ratio Handles small probability values for numerical stability Controls decision purity and complexity Measures model performance Preprocessin g PROPOSED ARCHITECTURE The model proposed will categorize people as Introverts or Extroverts through a machine learning-based system comprising preprocessing, feature engineering, and supervised classification methods. The pipeline model starts with preprocessing that involves the standardization of the numerical features to use StandardScaler, and the categorical features to use One-Hot Encoding to guarantee compatibility of the models. To establish a comparison of performances among the different ML algorithms (A), several MLA were used; these include LR, SVM, RF, XGBoost, NB and DT. Both of the models have been initially trained on untuned hyperparameters and then optimized to hyperparameter tuning using GridSearchCV with 5-fold CV. It was evaluated with the help of matrix perfromance. SVM was found to perform the best among the untuned models (92.93% accuracy) and when tuned Logistic Regression was the best model with a accuracy of 92.93% and equal precision-recall scores. The suggested solution shows strong classification performances and preprocessing and tuning in are relevant in improving prediction. PARAMETERS Performance Matrix V. RESULTS According to results of experiment, machine learning methods are effective in distinguishing between Introverts and Extroverts using their behavioral characteristics. First, six classifiers were tested in their untuned versions, namely, LR, SVM, RF, XGBoost, NB, and DT. SVM performance was the highest with the highest untuned accuracy of 92.93 percent 2 Authorized licensed use limited to: Chandigarh University. Downloaded on April 21,2026 at 09:24:50 UTC from IEEE Xplore. Restrictions apply. SVM, NB, and XGBoost, also improved to the same level of 92.93% accuracy after tuning. These results confirm the robustness of the dataset and the reliability of ML in predicting personality traits. closely followed by Naive bayes with the same accuracy and Decision tree with the lowest accuracy of 87.24 percent. Hyperparameter tuning with GridSearchCV was conducted. Interestingly, on tuning, the best-performing model was the Logistic Regression with a high accuracy of 92.93 and a precision of 0.94 and a recall of 0.92 and F1-score of 0.93. The similarity of performers among models indicates that there was a good organization of the dataset and sample. The analysis of the ROC curves showed that it was very separable with an AUC value of almost 0.95 across the models indicating strong robustness. Moreover, the visualization of confusion matrices revealed an equal distribution of classification of both classes of personalities. In general, the discussion confirms that hyperparameter tuning is a critical component of performance optimization, and Logistic Regression has accuracy and interpretability, which is the most appropriate model to use in this dataset. TABLE III. Class / Avg 0 (Introvert) 1 (Extrovert) B. Loss Indirectly in this work, loss was quantified in the number of misclassifications as most of the algorithms used were classical machine learning algorithms with cross-entropy loss or hinge loss functions implicit in training. The most successful models among the 580 samples of the test, namely; LR, SVM, NB, RF and XGBoost all had 92.93, which translates to around 41 samples not classified well and 539. After hyperparameter optimisation, Logistic Regression did not decline in performance, which suggests that there was very little overfitting and that the process of optimisation did not change substantially. Conversely, the Decision Tree model was the most misclassified with the highest loss of approximately 74 samples, with an accuracy of just 87.24%. These false groupings underscore the overfitting nature of the model because of its form. In general the low numbers of misclassification among tuned models indicate that the preprocessing stages such as scaling and encoding were successful to minimize variability and enhance generalization. The predictive ability of Logistic Regression with respect to minimizing its loss highlights its ability to generate reliable results on personality prediction problems. CLASSIFICATION REPORT Precision Recall F1-Score Support 94 92 93 302 92 94 93 278 93 580 Accuracy Macro Avg 93 93 93 580 Weighted Avg 93 93 93 580 A. Accuracy Fig. 3. Showing the graph of Training Loss & Validation Loss w.r.t Epochs C. Confusion Metrix Fig 2. Training and Validation Accuracy w.r.t Epochs The analysis of the proposed personality classification model showed that the accuracy is high in all the MLA. During the untuned stage, both SVM and NB classifier demonstrated the highest accuracy of 92.93, having been able to classify about 539 out of 580 test samples correctly. Logistic Regression and Random Forest followed closely with accuracies of 92.41% and 92.24%, while XGBoost achieved 91.72%, and Decision Tree performed the lowest with 87.24% accuracy. After applying GridSearchCV-based hyperparameter tuning, Logistic Regression appeared as the best-performing model with a tuned accuracy of 92.93%, maintaining a balance in both introvert and extrovert classifications. The performance of other models, such as RF, 3 Authorized licensed use limited to: Chandigarh University. Downloaded on April 21,2026 at 09:24:50 UTC from IEEE Xplore. Restrictions apply. The confusion matrix provides a clear visualization of the classification performance of the proposed model in distinguishing Introverts and Extroverts. For the bestperforming tuned model, Logistic Regression, the confusion matrix showed highly balanced results across both classes. The model was able to classify 278 out of 302 Introvert samples as Introverts, and falsely classify 24 samples as Extroverts. Equally, it was able to classify 260 out of 278 samples of Extroverts rightly and 18 wrongly as Introverts. This balance shows a good generalization of this model and its capability to manage the distribution of classes. The large percentage of correct predictions is brought out in the diagonal superiority of the matrix whereas the relative low off-diagonal values are reflective of low error levels. The total accuracy was 92.93, and the precision and recall scores are nearly equal 0.93 of each of the two classes, which means that the model is equally effective in reducing false positives and false negatives. The confusion matrix therefore confirms the strength of the Logistic Regression, and it is a good place to rely on when classifying personalities in behavioral data. VI. This paper revealed that machine learning methods are effective in personality prediction (that is differentiating between Introvert and Extrovert) along with behavioural data. Using preprocessing nodes like scaling and one-hot encoding, in addition to many types of classifiers, has provided the model with strong and stable results. With the untuned models, SVM had the highest accuracy of 92.93 but with the hyperparameter tuned, the best performing model was Logistic Regression, which had the same accuracy of 92.93 with a F1-score of 0.93. These results indicate that in addition to the fact that conventional models, such as Logistic Regression are reliable, hyperparameter optimization is necessary to enhance classification results. The equal performances in terms of accuracy, recall, and precision also prove the strength of the dataset and the offered pipeline. To expand its scope in the future, this research can be further developed with the introduction of deep learning structures, ensemble techniques and transfer learning techniques to be able to process larger and more varied data sets. Moreover, the incorporation of textual, physiological, or social media data may make the models more accurate and generalizable and open the way to real-life applications in recruitment, psychology, and personalized suggestions. Fig. 3. Showing the Confusion Matrix TABLE IV. COMPARISON RESEARCH APPROACHES. BETWEEN EXISTING Aspect Existing Research Dataset Size Often limited datasets with fewer behavioral features (typically <1000 samples). Preprocessing Models Used Hyperparameter Tuning Basic preprocessing; limited handling of categorical variables. Primarily Logistic Regression, Decision Tree, or Naïve Bayes. AND PROPOSED Proposed Research Large dataset of 2900 samples with 7 behavioral features for robust analysis. Comprehensive preprocessing using StandardScaler for numerical and OneHotEncoder for categorical features. Broader set of models including LR, SVM, RF, XGBoost, NB, and DT. Often ignored or limited grid search. Extensive GridSearchCVbased tuning with 5-fold crossvalidation for all models. Accuracy reported between 70–85% in earlier studies. Accuracy improved to 92.93% across tuned models, with F1-score = 0.93. REFERENCES [1] [2] [3] [4] [5] [6] Performance Metrics [7] Best Performing Model Logistic Regression or Decision Tree in smallscale studies. Generalization Moderate, with overfitting concerns in decision-based models. CONCLUSION & FUTURE SCOPE Logistic Regression (tuned) performed best, balancing accuracy and interpretability. Strong generalization with balanced classification confirmed via ROC-AUC ~0.95. [8] [9] Ahmad, S., Khan, R., & Malik, F. (2022). Personality prediction using machine learning approaches: A comparative analysis. Journal of Intelligent Systems, 31(4), 523–536. https://doi.org/10.1016/j.jintsys.2022.05.014 Sharma, N., & Gupta, R. (2022). Deep learning-based personality classification from text and behavioral data. Applied Soft Computing, 118, 108485. https://doi.org/10.1016/j.asoc.2022.108485 B. P. Joshi and S. Kumar, “A computational method of forecasting based on intuitionistic fuzzy sets and fuzzy time series,” in SocProS, New Delhi, India, Dec. 20–22, 2011, vol. 2, pp. 993–1000, doi: 10.1007/978-81-322-0491-6_91. W. A. Syed, J. Fang, V. S. B. Chilluri, M. Gu, N. P. Patil, and M. Kattimani, “Methods and systems for dynamic compression and transmission of application log data,” U.S. Patent 11 966 636 B2, Apr. 23, 2024. Available: https://patents.google.com/patent/US11966636B2/en. P. S. Pisal, J. Kishore, B. P. Joshi and S. Goyal, "Detection of Nanoparticles with Machine Learning Technique: Evaluation of Algorithm Performance," ICSIT, Nagpur, India, 2025, pp. 1-5, doi: 10.1109/ICSIT65336.2025.11295367. A. Sherov, T. Rakhimov, H. Hajiyev, M. Zelinskaya, and G. Khidirova, “Evaluating class-imbalanced data handling for enhanced financial distress prediction using an attention-based deep neural network and heuristic optimization algorithms,” Eng. Technol. Appl. Sci. Res., vol. 15, Art. no. 13372, 2025, doi: 10.48084/etasr.13372. A. Sherov, T. Rakhimov, H. Hajiyev, M. Zelinskaya, and G. Khidirova, “Assessment of class imbalance data handling with attention-based deep learning approach for robust financial distress prediction in enterprises,” Eng. Technol. Appl. Sci. Res., vol. 15, Art. no. 14843, 2025, doi: 10.48084/etasr.14843. I. Abdullayev, E. Akhmetshin, E. Hajiyev, Z. Mamadiyarov, and T. Khorolskaya, “A financial time series forecasting model using quasirecurrent neural networks and the crown porcupine optimizer for stock market risk prediction,” Eng. Technol. Appl. Sci. Res., vol. 15, Art. no. 13327, 2025, doi: 10.48084/etasr.13327. S. Aarthi, R. N. Ravikumar, M. Kalandarova, N. Khalikova, and E. Iskandarov, “Overcoming barriers in metaheuristic neural network optimization for biomedical imaging,” in Metaheuristic Algorithms and Optimizing Neural Networks for Biomedical Image Processing, IGI Global, 2025, doi: 10.4018/979-8-3373-0523-3.ch014. 4 Authorized licensed use limited to: Chandigarh University. Downloaded on April 21,2026 at 09:24:50 UTC from IEEE Xplore. Restrictions apply. [10] R. N. Ravikumar, S. Aarthi, B. S. Ruzimbaev, A. Satheesh Kumar, and M. Jumaniyozova, “AI-enhanced clinical decision-making through a collaborative approach,” in Applied AI and Computational Intelligence in Diagnostics and Decision-Making, IGI Global, 2025, doi: 10.4018/979-8-3373-3311-3.ch010. [11] Choudhary, Shilpa, Monali Gulhane, Sandeep Kumar, Nitin Rakesh, Sudhanshu Maurya, and Chanderdeep Tandon. "Integrating Machine Learning for Personalized Kidney Stone Risk Assessment: A Prospective Validation Using CLDN11 Genetic Data and Clinical Factors." Genomics at the Nexus of AI, Computer Vision, and Machine Learning (2025): 59-85. [12] Kumar, S., Sharma, K., Kumar, P. A., Jain, A., Bhagat, S. K., & Singh, P. (2024, September). An improved particle swarm approach for energy-aware location-aided routing in mobile ad-hoc network. In 2024 7th International Conference on Contemporary Computing and Informatics (IC3I) (Vol. 7, pp. 1119-1124). IEEE. [13] Verma P (2024) A Foodie’s Proselytization mediates Lifestyle and Affective Commitment: An application of Affect Heuristics in the hospitality sector. International Journal of Hospitality & Tourism Administration. 25(5) 875-895. DOI https://doi.org/10.1080/15256480.2023.2175288. [14] Patnaik, S., Wang, J. Y., Sadiq, F. U., & Sharma, K. (2025, December). Nutritional Interventions in Head and Neck Cancer Patients Undergoing Chemoradiotherapy: A Systematic Review and MetaAnalysis. In Healthcare (Vol. 13, No. 24, p. 3324). [15] Vadisetty, R. (2026). Bio-inspired AI Algorithms for Autonomous Agents: Revolutionizing Decision-Making, Resource Allocation, and Adaptability in Cloud Networks Through Nature-Inspired Models. In: Swaroop, A., Virdee, B., Correia, S.D., Polkowski, Z. (eds) Proceedings of Data Analytics and Management. ICDAM 2025. Lecture Notes in Networks and Systems, vol 1600. Springer, Cham. https://doi.org/10.1007/978-3-032-03072-6_40 [16] Bansal, Shonak, Sandeep Kumar, Arpit Jain, Vinita Rohilla, Krishna Prakash, Anupma Gupta, Tanweer Ali et al. "Design and TCAD analysis of few-layer graphene/ZnO nanowires heterojunction-based photodetector in UV spectral region." Scientific Reports 15, no. 1 (2025): 7762. [17] Singh, G., Sharma, S., Dhanny, B. H. S., & Garg, V. (Eds.). (2024). HR 4.0 Practices in the Post-COVID-19 Scenario. CRC Press.. [18] Hareesh, B., Moses C. John, and MVV Prasad Kantipudi. "VLSI Architectures of Booth Multiplication Algorithms--A Review." International Journal of Computing and Digital Systems 11, no. 1 (2022): 265-276. [19] Jonnala, Naga Surekha, Renuka Chowdary Bheemana, Krishna Prakash, Shonak Bansal, Arpit Jain, Vaibhav Pandey, Mohammad Rashed Iqbal Faruque, and K. S. Al-Mugren. "DSIA U-Net: deep shallow interaction with attention mechanism UNet for remote sensing satellite images." Scientific Reports 15, no. 1 (2025): 549. [20] Singh, A., Luthra, A., Garg, S., & Sharma, V. (2025). Metaverse Adoption Among Banking Users: A Developing Nation's Perspective. In The AI Metaverse Revolution: Transforming Multi-business Scenarios (Volume 1) (pp. 95-114). Emerald Publishing Limite. [21] Vadisetty, R., Polamarasetti, A., Varadarajan, V., Kalla, D., Ramanathan, G.K. (2026). Cyber Warfare and AI Agents: Strengthening National Security Against Advanced Persistent Threats (APTs). In: Dhoska, K., Spaho, E. (eds) AI and Digital Transformation: Opportunities, Challenges, and Emerging Threats in Technology, Business, and Security. ICITTBT 2025. Communications in Computer and Information Science, vol 2669. Springer, Cham. https://doi.org/10.1007/978-3-032-07373-0_43. 5 Authorized licensed use limited to: Chandigarh University. Downloaded on April 21,2026 at 09:24:50 UTC from IEEE Xplore. Restrictions apply.