An Integrated Feature Extraction Based on Principal Components and Deep Auto Encoder with Extra Tree for Intrusion Detection Systems

Abstract

With advances in computer network technology, the Internet has become an integral part of our daily lives. It goes without saying that providing security to network data is crucial. To this effect, the intrusion detection system (IDS) is vital for defending networks against cyberattacks. However, the high dimensionality of data is a significant issue that impacts categorisation accuracy. Therefore, we proposed a novel integrated feature extraction method called PC-UDAE that uses principal component analysis (PCA) and Unsupervised Deep Autoencoder (UDAE) to extract linear and nonlinear relationships. Also, to address the class imbalance, synthetic minority class samples are generated using the combination of Synthetic Minority Over Sampling Technique and Edited Nearest Neighbour (SMOTE-ENN). Finally, the extracted features are trained by using supervised machine learning models like Random Forest (RF), Extreme Gradient Boosting Machine (XGBM), Decision Tree (DT), Light Gradient Boosting Machine (LGBM), Extra Tree (ET), Support Vector Machine (SVM), AdaBoost (AB), and K-Nearest Neighbour (KNN) with the original imbalanced and balanced data. We analysed our suggested model using UNSW-NB 15, NSL-KDD, Ton-IoT data sets and obtained 98.85%, 99.59%, and 99.97% accuracy, respectively. Our experimental findings show that our proposed method outperformed all other competing methods.