A Comparative Analysis on Ensemble Learning and Deep Learning Based Intrusion Detection Systems over the NCC2 Dataset

Intrusion detection systems (IDS) are effective countermeasures to ensure Internet of Things (IoT) network security. IDS are employed to guarantee data privacy and assess network integrity through the detection of malicious activities at an early stage. IDS aims to accurately characterize normal traffic pattern and identify behavior deviation as malicious. This paper presents a comparative study on advanced machine learning (ML) and deep learning (DL) techniques for attack traffic detection and type identification over simultaneous parallel sensors. Performances of four classifiers including Extreme Gradient Boosting (XG-Boost), CNN-LSTM, Autoencoders (EA), and Multi-layer Perceptron (MLP) for multiclass classification problem to accurately identify distributed cyber-attacks type over the NCC2 dataset. Experimental results reveal that the XGBoost classifier over the RF based feature set produces an accuracy and precision up to 1 outperforms other classifiers. Comparative analysis indicates the importance of feature extraction techniques and their ability to affect classifier performances.