Selected work

Representative Papers

Prediction of bank transaction fraud using TabNet an adaptive deep learning architecture

Jul 20, 2026

This study addresses the critical need for high accuracy, interpretability, and scalability in online banking fraud detection by leveraging real-world transaction data from India. Through exploratory data analysis and SMOTE-based oversampling to mitigate class imbalance, the authors systematically evaluate five deep learning models: DNN, GRU, LSTM, 1D-CNN, and TabNet. Notably, they harness TabNet’s intrinsic sparse feature selection mechanism to simultaneously enhance model interpretability and generalization. Experimental results demonstrate that TabNet achieves a 97.39% accuracy and a 0.9739 ROC-AUC under three-fold cross-validation, significantly outperforming baseline models. The approach effectively reduces both false positives and false negatives, supports real-time deployment, and satisfies stringent financial regulatory requirements for model transparency.

0 citationsRead paper

Predictive Analytics in E-Commerce for CustomerBehavior Forecasting using hybrid Ret-DNN withXGBoost Model

Jun 16, 2026

This study addresses the challenge of accurately predicting users’ future purchasing behavior on e-commerce platforms. To this end, the authors propose a novel hybrid model that integrates a Recurrent Temporal Deep Neural Network (Ret-DNN) with XGBoost. Specifically, Ret-DNN is employed for the first time as a sequential feature extractor to learn deep representations from users’ transaction sequences, and its outputs are subsequently fed into XGBoost to predict purchase probabilities. This architecture effectively combines the strengths of deep learning in capturing temporal dependencies with the predictive power of gradient-boosted decision trees. Evaluated on a real-world UK e-commerce dataset comprising 500,000 records, the proposed model achieves a mean absolute error (MAE) of 0.2193, significantly outperforming the standalone Ret-DNN and demonstrating both the effectiveness and innovation of the approach in customer behavior prediction.

0 citationsRead paper

Cyber Threat Detection and Vulnerability Assessment System Using Generative AI and Large Language Model

Oct 17, 20252025 2nd International Conference on Software, Systems and Information Technology (SSITCON)

This work proposes a RoBERTa-based system for network threat detection and vulnerability assessment that addresses the limitations of existing security models—such as Security BERT—in identifying sophisticated cyberattacks like ransomware, phishing, and DoS within encrypted traffic due to insufficient contextual understanding. The approach extracts traffic data from PCAP files, preserves privacy through fully homomorphic encryption (FHE), and employs a byte-level Byte Pair Encoding (BBPE) tokenizer to generate semantically rich tokens for input into the RoBERTa model. Experimental results demonstrate significant improvements over current BERT-based models, achieving an accuracy of 0.99, recall of 0.91, and precision of 0.89, thereby enhancing semantic comprehension and detection efficacy for advanced threats in encrypted network traffic.

0 citationsRead paper

Recent publications

Latest Papers

Prediction of bank transaction fraud using TabNet an adaptive deep learning architecture

Jul 20, 2026

This study addresses the critical need for high accuracy, interpretability, and scalability in online banking fraud detection by leveraging real-world transaction data from India. Through exploratory data analysis and SMOTE-based oversampling to mitigate class imbalance, the authors systematically evaluate five deep learning models: DNN, GRU, LSTM, 1D-CNN, and TabNet. Notably, they harness TabNet’s intrinsic sparse feature selection mechanism to simultaneously enhance model interpretability and generalization. Experimental results demonstrate that TabNet achieves a 97.39% accuracy and a 0.9739 ROC-AUC under three-fold cross-validation, significantly outperforming baseline models. The approach effectively reduces both false positives and false negatives, supports real-time deployment, and satisfies stringent financial regulatory requirements for model transparency.

0 citationsRead paper

Predictive Analytics in E-Commerce for CustomerBehavior Forecasting using hybrid Ret-DNN withXGBoost Model

Jun 16, 2026

This study addresses the challenge of accurately predicting users’ future purchasing behavior on e-commerce platforms. To this end, the authors propose a novel hybrid model that integrates a Recurrent Temporal Deep Neural Network (Ret-DNN) with XGBoost. Specifically, Ret-DNN is employed for the first time as a sequential feature extractor to learn deep representations from users’ transaction sequences, and its outputs are subsequently fed into XGBoost to predict purchase probabilities. This architecture effectively combines the strengths of deep learning in capturing temporal dependencies with the predictive power of gradient-boosted decision trees. Evaluated on a real-world UK e-commerce dataset comprising 500,000 records, the proposed model achieves a mean absolute error (MAE) of 0.2193, significantly outperforming the standalone Ret-DNN and demonstrating both the effectiveness and innovation of the approach in customer behavior prediction.

0 citationsRead paper

Cyber Threat Detection and Vulnerability Assessment System Using Generative AI and Large Language Model

Oct 17, 20252025 2nd International Conference on Software, Systems and Information Technology (SSITCON)

This work proposes a RoBERTa-based system for network threat detection and vulnerability assessment that addresses the limitations of existing security models—such as Security BERT—in identifying sophisticated cyberattacks like ransomware, phishing, and DoS within encrypted traffic due to insufficient contextual understanding. The approach extracts traffic data from PCAP files, preserves privacy through fully homomorphic encryption (FHE), and employs a byte-level Byte Pair Encoding (BBPE) tokenizer to generate semantically rich tokens for input into the RoBERTa model. Experimental results demonstrate significant improvements over current BERT-based models, achieving an accuracy of 0.99, recall of 0.91, and precision of 0.89, thereby enhancing semantic comprehension and detection efficacy for advanced threats in encrypted network traffic.

0 citationsRead paper