Publications

This work presents a stacked LSTM intrusion detection system tailored for IoT environments using the NSL-KDD dataset. The approach combines chi-square based feature engineering with deep sequence modeling to raise precision, recall, and overall robustness against DoS, intrusions, and data leakage. The resulting model strengthens IoT infrastructure security while remaining scalable for large deployments.

The study introduces an AI and machine learning powered automated incident response system (AIRS) for IoT networks. By enabling self-healing network routines, the framework cuts mean time to respond, scales across smart city and industrial IoT scenarios, and lays the groundwork for future extensions with federated learning and privacy preserving models.

This article details a federated learning based intrusion detection architecture that reinforces confidentiality through SSL and TLS channels. The approach reduces communication overhead, defends against stealthy false data injection attacks, and demonstrates strong performance on datasets such as CIFAR10 and MNIST. The framework advances privacy aware threat mitigation for IoT and edge networks.

The paper evaluates multiple convolutional neural network pipelines for MRI brain tumor segmentation, highlighting ensemble U-Net models that blend 3D and 2D processing. A comprehensive metric analysis shows significant improvements in diagnostic accuracy, informing personalized treatment planning and AI driven clinical workflows.

This research delivers a CNN powered decision support engine for crop recommendation that interprets soil chemistry, temperature, humidity, pH, and rainfall indicators. By embracing explainable AI principles, the method elevates precision in agricultural planning and supports sustainable yield outcomes.

The study benchmarks neural networks, Gaussian naive Bayes, support vector machines, and decision trees on generated IoT datasets against KDDCUP99. Feature selection distills only five key attributes, cutting execution time while enhancing accuracy. An improved CNNwGFC model further elevates classification effectiveness on UNSW-NB15.

This research surveys Internet of Things performance metrics published between 2010 and 2021, distilling 158 measures into twelve thematic categories. The resulting ontology highlights the predominance of network metrics and guides practitioners toward improved IoT performance management and optimization.