Feature extraction for machine learning-based intrusion detection in IoT networks

A large number of network security breaches in IoT networks have demonstrated the unreliability current Network Intrusion Detection Systems (NIDSs). Consequently, interruptions and loss sensitive data occurred, which led to an active research area for improving NIDS technologies. In analysis related works, it was observed that most researchers aim obtain better classification results by using a set untried combinations Feature Reduction (FR) Machine Learning (ML) techniques on datasets. However, these datasets are different feature sets, attack types, design. Therefore, this paper aims discover whether can be generalised across various Six ML models utilised: Deep Feed Forward (DFF), Convolutional Neural (CNN), Recurrent (RNN), Decision Tree (DT), Logistic Regression (LR), Naive Bayes (NB). The accuracy three Extraction (FE) algorithms; Principal Component Analysis (PCA), Auto-encoder (AE), Linear Discriminant (LDA), evaluated benchmark datasets: UNSW-NB15, ToN-IoT CSE-CIC-IDS2018. Although PCA AE algorithms been widely used, determination their optimal extracted dimensions has overlooked. indicate no clear FE method or model achieve best scores all identified each dataset, LDA degrades performance two variance is used analyse PCA. Finally, concludes choice significantly alters applied techniques. We believe universal (benchmark) needed facilitate further advancement progress field.