Exploring feasible malicious attack detection in IoT enabled network through metaheuristic and deep learning approach

The increasing integration of Internet of Things (IoT) devices into critical infrastructure has expanded the attack surface, making network security a paramount concern. While intrusion detection systems (IDS) are essential for protecting these networks, traditional methods often struggle with the high-dimensional nature of IoT traffic, failing to extract the most relevant features efficiently and suffering from high false positive rates. This study aims to develop a robust and adaptable deep learning-based intrusion detection framework specifically designed to overcome these limitations by enhancing feature selection and model optimization for improved threat identification in IoT environments. To achieve this, the proposed methodology first employs Minimum Redundancy Maximum Relevance (mRMR) to preprocess data and select the most significant features, reducing noise and redundancy. Subsequently, hybrid deep learning models integrating Convolutional Neural Networks (CNN) and Deep Neural Networks (DNN) with Long Short-Term Memory (LSTM) networks are developed. The performance of these hybrid models is further optimized by fine-tuning their hyperparameters using the Grasshopper Optimization Algorithm, resulting in two novel models: mrCNN-LSTM-IDS and mrDNN-LSTM-IDS. The experimental results, derived from comprehensive testing on the TON-IoT and NSL-KDD datasets, demonstrate that the mrDNN-LSTM-IDS model significantly outperforms both its CNN-based counterpart and other individual deep learning models. It achieved exceptional performance metrics, including an accuracy of 99.89% and 99.86%, precision of 99.47% and 99.89%, recall of 99.85% and 99.64%, an F1-score of 99.45% and 99.21%, and a remarkably low false positive rate of 0.0002 and 0.0001 on NSL-KDD and ToN -IoT datasets respectively. This study concludes that the proposed mrDNN-LSTM-IDS framework offers a highly effective and adaptable solution for real-world IoT network security, successfully addressing key challenges in feature extraction and attack detection accuracy.