Improving the Performance of Network Intrusion Detection Systems in IoT Using Grid Search & Feature Selection by Decision Tree

Neural networks like deep learning have become effective solutions for NIDS. Their ability to learn intricate patterns and behaviors makes them well-suited for distinguishing between normal network traffic and attacks. However, a significant drawback lies in the resource-intensive training process. Many network gateways and routers lack the necessary memory and processing power to train and execute such models. This study leverages the Botnet Malware attacks of the Kitsune dataset. Our analysis focuses on evaluating this dataset using various algorithms in an offline mode, which proves to be more efficient and exact for detecting attacks within local networks. The proposed method for feature selection uses the efficiency detector value (EDV). The decision tree has the best results between deep learning, linear regression, and random forest algorithms with 99.97 % recall, which has 0.28 % improvement versus deep federated learning. The proposed feature selection method improves the runtime by about 67 % using the proposed feature selection method.