Research

A comparative study evaluating Vision Transformers (ViT-B32) and EfficientNet-B0 for post-disaster building damage assessment using Hurricane Harvey satellite imagery. The ViT model achieves higher accuracy (97.85%) and demonstrates superior robustness on imbalanced data, while attention heatmaps provide interpretable localization of damaged structures. 

A machine learning–based study for stress detection using Photoplethysmography (PPG) signals. The framework combines signal preprocessing, feature engineering, and supervised classification techniques to identify stress levels from physiological data. Experimental results demonstrate outstanding performance, with the Gradient Boosting Classifier achieving 99.77% accuracy, highlighting the potential of wearable sensing and AI-driven health monitoring systems. 

A machine learning–based study for stress classification using physiological signal data, where multiple classifiers including KNN, SVM, Naive Bayes, and AdaBoost are evaluated against a proposed Gradient Boosting Classifier (GBC). After signal preprocessing, feature extraction, and supervised training, the GBC model outperforms all baseline methods, achieving high classification accuracy (99.77%) along with strong precision, recall, and F1-score. The results demonstrate robust differentiation between stressed and non-stressed states, highlighting the framework’s suitability for reliable real-time stress detection.