Cross Model Verification of Intrusion Detection System on IoT Using Convolutional Neural Network

The IDS to secure IoT devices have become so important with the tremendous growth in the field of IoT. The potential damage to the IoT network by different types of attacks is increasing rapidly with the increased use of IoT devices. Therefore, IDS has become an essential defense tool against the vulnerable attacks. These IoT devices are to be continuously monitored for attacks, hence IDS software is deployed to detect anomalies. Today ML and DL algorithms which are subset of Artificial Intelligence are widely used in the intrusion detection system to detect attacks. In this research work, the cross verification of the outputs of IDS using CNN on three different datasets namely NSLKDD, IDS2018 and IOTID20 are explored. Intrusion detection in the context of the IoT has become a critical concern due to the proliferation of connected devices and evolving attack techniques. Traditional IDS often struggle to adapt to the unique challenges posed by IoT environments. This article explores the application of CNNs for intrusion detection in IoT and discusses the significance of cross-model verification to enhance security and accuracy. The AUC score yielded an accuracy of more than 80-95% in 15 epochs in all the three datasets. Both binary and multiclassification are done. Further it is observed that the ROC curve is above the diagonal line signifying the model is excellent. Moreover, a new feature selection algorithm SELBEST is proposed based on Stochastic gradient descent, which significantly reduced the features. Also, the classification time is reduced and the performance is increased.