Publications

A novel framework for malware detection utilizing entropy-based statistical features and machine learning models across different file types. The approach provides effective detection capabilities for various malware variants.

A hybrid approach combining ensemble models with adaptive synthetic sampling for malicious URL detection. The method improves detection accuracy and handles class imbalance effectively in cybersecurity applications.

Analyzing the impact of recursive feature elimination and information gain techniques on machine learning models for student performance prediction. The research highlights the importance of feature selection in educational data mining.

OPTISTACK presents a hybrid ensemble learning approach combined with explainable AI for malware detection in compressed files. The framework provides both high detection accuracy and interpretability for security analysts.

A hybrid intrusion detection system combining Principal Component Analysis (PCA) and SMOTE with advanced ensemble models. The approach enhances network security through improved detection of various intrusion patterns.

A novel hybrid ensemble model combining Random Forest, XGBoost, and K-Nearest Neighbors with probabilistic fusion and meta-level calibration. RXK-VEM outperforms traditional classifiers in F1 score, AUC-ROC, and MCC, especially improving recall for underrepresented classes in imbalanced educational datasets.

Investigates the impact of Adaptive Synthetic Sampling (ADASYN) on breast cancer classification across diverse ML models. Results show significant recall improvements for KNN (90.5% to 95.2%) and XGBoost (88.1% to 95.2%), demonstrating ADASYN's effectiveness in enhancing sensitivity to minority-class cases.

A stacked ensemble model integrating five heterogeneous classifiers with logistic regression as meta-learner. The framework incorporates Z-score normalization and SMOTE, achieving AUC of 0.711 and recall of 0.748. SHAP-based interpretability reveals key contributing features for clinical adoption.

A lightweight CNN model for efficient and interpretable white blood cell classification, achieving 84% accuracy across four WBC types. Grad-CAM and Grad-CAM++ visualizations provide class-specific attention maps aligned with morphological features, making it suitable for real-time diagnostic support.

An integrated framework combining SMOTE, correlation-based feature selection, and explainable AI for multiclass student performance prediction. Random Forest achieves perfect classification, with XAI methods (SHAP, LIME) identifying project and assignment scores as most influential predictors.

AWV-ECC, an adaptive weighted voting ensemble that weights classifiers by error rates and penalizes correlated misclassifications. Evaluated on 58,596 memory dumps, achieves near-perfect performance (accuracy 0.9999, AUC 1.0000), providing robust detection of obfuscated malware.

Empirical evidence on when to use standardization and normalization in machine learning models with explainable AI insights. This study provides comprehensive analysis of data preprocessing techniques and their impact on model performance and interpretability.

Investigating the effects of imbalanced datasets on machine learning algorithms in the context of predicting student performance. The study explores various techniques to handle class imbalance and improve prediction accuracy.

A comparative analysis of ensemble and linear models using Synthetic Minority Over-sampling Technique (SMOTE) for student performance prediction. The study demonstrates the effectiveness of SMOTE in handling imbalanced educational datasets.