Abstract
Intrusion detection systems (IDS) protect networks by continuously monitoring data flow and taking immediate action when anomalies are detected. However, due to redundancy and significant network data correlation, classical IDS have shortcomings such as poor detection rates and high computational complexity. This paper proposes a novel feature selection and extraction technique (FI-PCA). Feature Importance (FI) and Principal Component Analysis (PCA) are used to preprocess the network dataset (PCA). FI identifies the most important features in the data, while PCA is used to reduce dimensionality and denoise the data. In order to detect anomalies, we employ three single classifiers: Decision Tree (DT), Naive Bayes and Logistic Regression. Preliminary results, however, show that these classifiers have achieved average classification metric scores. On this basis, we use the Stack Ensemble Learning Classifier (ELC) method of combining single classifiers to improve the classifier's performance further. Experimental results on varied feature dimensions of an IoT (Bot-IoT) dataset indicate that our proposed technique combined with the Stack ELC can maintain the same level of classification performance for reduced dataset features. A comparison of our result with state-of-the-art classifiers' classification performance shows that our classifier is superior in terms of accuracy and detection rate. At the same time, a remarkable decrease is recorded for both training and test time.