Shital Parag Dongre
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Shital Parag Dongre has a total of 18 co-authors. Here are the top co-authors by number of shared publications:
- Sharanyaa Sigamani
Panimalar Engineering College
2 shared publications - Aayusha Marne
Vishwakarma Institute of Technology,AI&DS,India
1 shared publication - Ayesha Pathan
Vishwakarma Institute of Technology,AI&DS,India
1 shared publication - Bharanidharan Gurumurthy
The New College
1 shared publication - Durga Devi Saravanan
Vel Tech High Tech Dr. Rangarajan Dr. Sakunthala Engineering College
1 shared publication - Hema Mayavan
Assistant Professor, Department of Information Technology Easwari Engineering College Ramapuram, Chennai, Tamil Nadu, India
1 shared publication
Latest Publications
- Open Access
- 1 December 2026
- Published by Institute of Advanced Engineering and Science in International Journal of Informatics and Communication Technology (IJ-ICT)
- Vol. 15 (4) , 1816-1828
AI-powered framework for advancing 5G and 6G wireless communication
A shift to 5G to 6G networks would provide better connectivity, transmission capacity, security, and latency, but there are still challenges such as dynamic resource allocation, energy consumption, network optimisation, and cybersecurity issues. This research presents an artificial intelligence (AI)-based approach to enhance the accuracy, efficiency and security of 5G and future 6G wireless networks. Adaptive spectrum and resource allocation is done using deep reinforcement learning (DRL), and AI traffic forecasting and dynamic beamforming helps reduce latency, boost throughput and ensure quality of service (QoS). Federated learning (FL) trains models at the distributed sites while minimizing the amount of raw data transmission, and generative adversarial networks (GANs) create representative samples of data for network and cybersecurity analysis. The anomaly detection feature powered by AI, along with the blockchain-based authentication, helps enhance access control and threat detection. The framework was tested and analyzed in a simulation environment built in Python and in comparison to traditional optimization. It performed an F1-score of 0.94, versus 0.82; AUC–ROC of 0.96, versus 0.86; and resource-allocation success rate (91% versus 75%). These results represent the relative increases of 14.6%, 11.6% and 21.3%. The results suggest that AI-driven coordinated optimisation could lead to a more adaptive management of resources, threat detection, and efficient operation of networks compared with traditional methods, paving the way for intelligence, security and energy-efficient 6G communication systems.
- 1 December 2026
- Published by Institute of Advanced Engineering and Science in International Journal of Informatics and Communication Technology (IJ-ICT)
- Vol. 15 (4) , 1816-1828
Open AccessAI-powered framework for advancing 5G and 6G wireless communication
A shift to 5G to 6G networks would provide better connectivity, transmission capacity, security, and latency, but there are still challenges such as dynamic resource allocation, energy consumption, network optimisation, and cybersecurity issues. This research presents an artificial intelligence (AI)-based approach to enhance the accuracy, efficiency and security of 5G and future 6G wireless networks. Adaptive spectrum and resource allocation is done using deep reinforcement learning (DRL), and AI traffic forecasting and dynamic beamforming helps reduce latency, boost throughput and ensure quality of service (QoS). Federated learning (FL) trains models at the distributed sites while minimizing the amount of raw data transmission, and generative adversarial networks (GANs) create representative samples of data for network and cybersecurity analysis. The anomaly detection feature powered by AI, along with the blockchain-based authentication, helps enhance access control and threat detection. The framework was tested and analyzed in a simulation environment built in Python and in comparison to traditional optimization. It performed an F1-score of 0.94, versus 0.82; AUC–ROC of 0.96, versus 0.86; and resource-allocation success rate (91% versus 75%). These results represent the relative increases of 14.6%, 11.6% and 21.3%. The results suggest that AI-driven coordinated optimisation could lead to a more adaptive management of resources, threat detection, and efficient operation of networks compared with traditional methods, paving the way for intelligence, security and energy-efficient 6G communication systems.
- Open Access
- 30 September 2026
- Published by VFAST Research Platform in VFAST Transactions on Software Engineering
- Vol. 14 (3) , 628-643
A Hybrid Deep Reinforcement Learning Framework for Enhanced Cyber Threat Detection Using Supervised Ensemble Decision Fusion
Conventional intrusion detection systems (IDSs) face increasing complexity in network traffic and unpredictable cyberattacks, particularly with fixed supervised classifiers and ambiguous observations. This study introduces DRL-ECTD (Deep Reinforcement Learning-Enhanced Cyber Threat Detector), a hybrid binary IDS that combines probability estimates from Random Forest (RF) and Artificial Neural Network (ANN) classifiers with a cost-sensitive decision-refinement mechanism using Deep Q-Network (DQN). High-confidence predictions are retained, while predictions within a preset uncertainty interval are passed to the DQN for secondary evaluation. Individual NSL-KDD records are treated as independent decision states, with DQN used for reinforcement-based refinement rather than temporal network-state modeling. On the NSL-KDD test set, DRL-ECTD achieved 78.6% accuracy, 80.4% precision, 76.3% recall, 78.3% F1-score, and 0.988 AUC. The ANN baseline achieved slightly higher accuracy (79.2%) and F1 score (79.1%). DRL-ECTD outperforms the other tested models in accuracy and F1-score and the baseline RF and standalone DQN in precision. SHAP was used for feature attribution and model interpretation. A proof-of-concept Flask service interface demonstrated deployment feasibility, although latency, throughput, computational overhead, and resource utilization were not experimentally benchmarked. The results provide initial insights into hybrid confidence-based decision refinement and motivate further experiments with multiple classes, temporal order, and deployment for intrusion detection.
- 30 September 2026
- Published by VFAST Research Platform in VFAST Transactions on Software Engineering
- Vol. 14 (3) , 628-643
Open AccessA Hybrid Deep Reinforcement Learning Framework for Enhanced Cyber Threat Detection Using Supervised Ensemble Decision Fusion
Conventional intrusion detection systems (IDSs) face increasing complexity in network traffic and unpredictable cyberattacks, particularly with fixed supervised classifiers and ambiguous observations. This study introduces DRL-ECTD (Deep Reinforcement Learning-Enhanced Cyber Threat Detector), a hybrid binary IDS that combines probability estimates from Random Forest (RF) and Artificial Neural Network (ANN) classifiers with a cost-sensitive decision-refinement mechanism using Deep Q-Network (DQN). High-confidence predictions are retained, while predictions within a preset uncertainty interval are passed to the DQN for secondary evaluation. Individual NSL-KDD records are treated as independent decision states, with DQN used for reinforcement-based refinement rather than temporal network-state modeling. On the NSL-KDD test set, DRL-ECTD achieved 78.6% accuracy, 80.4% precision, 76.3% recall, 78.3% F1-score, and 0.988 AUC. The ANN baseline achieved slightly higher accuracy (79.2%) and F1 score (79.1%). DRL-ECTD outperforms the other tested models in accuracy and F1-score and the baseline RF and standalone DQN in precision. SHAP was used for feature attribution and model interpretation. A proof-of-concept Flask service interface demonstrated deployment feasibility, although latency, throughput, computational overhead, and resource utilization were not experimentally benchmarked. The results provide initial insights into hybrid confidence-based decision refinement and motivate further experiments with multiple classes, temporal order, and deployment for intrusion detection.
- Open Access
- 9 July 2026
- Published by VFAST Research Platform in VFAST Transactions on Software Engineering
- Vol. 14 (3) , 569-590
GHO-Net: An Optimized and Explainable LightGBM Framework for 5G Core Network Anomaly Detection
The high dimensionality of traffic, severe class imbalance, hyperparameter sensitivity, and lack of model interpretability pose challenges to reliable multiclass anomaly detection in Fifth-Generation (5G) core networks. The study presents an optimization-guided and explainable machine-learning framework, Grey Wolf Hyperparameter-Optimized LightGBM Network (GHO-Net), incorporating traffic-flow preprocessing, six domain-derived features, Mutual Information (MI) feature selection, training-only Random Oversampling (ROS), Grey Wolf Optimization (GWO), Light Gradient Boosting Machine (LightGBM) classification, and SHapley Additive exPlanations (SHAP) feature attribution. A total of 1,212,532 traffic records were distributed among nine classes in the cleaned 5G Network Intrusion Detection Dataset (5G-NIDD). MI selected 35 of 53 candidate predictors based on the validation weighted F1-score, and GWO optimized seven LightGBM hyperparameters. GHO-Net achieved 98.1885% accuracy, 98.2100% weighted precision, 98.1885% recall, 98.1887% weighted F1-score, and 99.6784% specificity on an independent test partition. It also obtained ROC-AUC of 0.998 and PR-AUC of 0.985. Pairwise comparisons showed that GHO-Net achieved higher accuracy on hard labels, while Decision Tree, Random Forest, and baseline LightGBM recorded slightly higher accuracy. Baseline LightGBM has the highest ROC-AUC and PR-AUC, whereas GHO-Net has the lowest log loss. The measured inference latency was 0.176860 ms per flow in the offline CPU benchmark. Due to this latency, which is higher than that of the baseline classifiers, and testing in live deployment, it cannot be considered suitable for real-time use in 5G core monitoring.
- 9 July 2026
- Published by VFAST Research Platform in VFAST Transactions on Software Engineering
- Vol. 14 (3) , 569-590
Open AccessGHO-Net: An Optimized and Explainable LightGBM Framework for 5G Core Network Anomaly Detection
The high dimensionality of traffic, severe class imbalance, hyperparameter sensitivity, and lack of model interpretability pose challenges to reliable multiclass anomaly detection in Fifth-Generation (5G) core networks. The study presents an optimization-guided and explainable machine-learning framework, Grey Wolf Hyperparameter-Optimized LightGBM Network (GHO-Net), incorporating traffic-flow preprocessing, six domain-derived features, Mutual Information (MI) feature selection, training-only Random Oversampling (ROS), Grey Wolf Optimization (GWO), Light Gradient Boosting Machine (LightGBM) classification, and SHapley Additive exPlanations (SHAP) feature attribution. A total of 1,212,532 traffic records were distributed among nine classes in the cleaned 5G Network Intrusion Detection Dataset (5G-NIDD). MI selected 35 of 53 candidate predictors based on the validation weighted F1-score, and GWO optimized seven LightGBM hyperparameters. GHO-Net achieved 98.1885% accuracy, 98.2100% weighted precision, 98.1885% recall, 98.1887% weighted F1-score, and 99.6784% specificity on an independent test partition. It also obtained ROC-AUC of 0.998 and PR-AUC of 0.985. Pairwise comparisons showed that GHO-Net achieved higher accuracy on hard labels, while Decision Tree, Random Forest, and baseline LightGBM recorded slightly higher accuracy. Baseline LightGBM has the highest ROC-AUC and PR-AUC, whereas GHO-Net has the lowest log loss. The measured inference latency was 0.176860 ms per flow in the offline CPU benchmark. Due to this latency, which is higher than that of the baseline classifiers, and testing in live deployment, it cannot be considered suitable for real-time use in 5G core monitoring.
- 6 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1505-1512
An Intelligent Gesture-Based Assistive Framework for Smart Presentations and Human-Computer Interaction
A smart gesture-based assistive platform to improve smart presentations and human-computer interface with physically challenged users. The proposed system offers an artificial intelligence-based and touchless presentation control system which allows controlling the slides as well as zooming images on the screen and writing annotations on the screen in real-time with the help of hand-gestures recognition. A streamlined MediaPipe-OpenCV system can be used to achieve a powerful hand tracking and adaptive gesture filtering and misclassification correction can make the system more reliable in the face of diverse user actions and environment parameters. The structure exhibits good adaptability to the environment and it has a high recognition performance in the case of low-light conditions and partial hand covering. Experimental analysis gives an average accuracy of 95, which is higher than baseline gesture recognition methods in all the conditions evaluated. The work offers a scaleable, user-friendly interface that promotes assistive technology and further develops practical gesture-driven interaction among educators, presenters, and other people with motor deficiencies, thus facilitating the creation of more inclusive digital interactions. With the integration of MediaPipe and OpenCV, gesture recognition is on the rise, and it is opening up the possibilities in the field of digital education, interaction, and assistive technologies. It is this inherent aspect of consideration of multi-user adjustability and the expansive variety of gestures that will form the basis of practicality and the expected large variety of applications in education, business, and artistic contexts. The findings suggest that the system offers a non-discriminatory, friendly, and interactive modality of interaction that minimizes reliance on hardware.
- 6 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1505-1512
An Intelligent Gesture-Based Assistive Framework for Smart Presentations and Human-Computer Interaction
A smart gesture-based assistive platform to improve smart presentations and human-computer interface with physically challenged users. The proposed system offers an artificial intelligence-based and touchless presentation control system which allows controlling the slides as well as zooming images on the screen and writing annotations on the screen in real-time with the help of hand-gestures recognition. A streamlined MediaPipe-OpenCV system can be used to achieve a powerful hand tracking and adaptive gesture filtering and misclassification correction can make the system more reliable in the face of diverse user actions and environment parameters. The structure exhibits good adaptability to the environment and it has a high recognition performance in the case of low-light conditions and partial hand covering. Experimental analysis gives an average accuracy of 95, which is higher than baseline gesture recognition methods in all the conditions evaluated. The work offers a scaleable, user-friendly interface that promotes assistive technology and further develops practical gesture-driven interaction among educators, presenters, and other people with motor deficiencies, thus facilitating the creation of more inclusive digital interactions. With the integration of MediaPipe and OpenCV, gesture recognition is on the rise, and it is opening up the possibilities in the field of digital education, interaction, and assistive technologies. It is this inherent aspect of consideration of multi-user adjustability and the expansive variety of gestures that will form the basis of practicality and the expected large variety of applications in education, business, and artistic contexts. The findings suggest that the system offers a non-discriminatory, friendly, and interactive modality of interaction that minimizes reliance on hardware.