A Machine Learning Approach for Intrusion Detection System on NSL-KDD Dataset

In the field of Network Security, there is a constant expedition towards the cyber-attacks which can lead to a destabilized network. Moreover, with unexpected inception and increased use of the Internet, malicious activities in the network are at a rapid upsurge. To detect anomalies in the network, it is crucial to construct a robust intrusion detection system (IDS) that combats unauthorized access to network resources so that information can be secured. Lately, various noteworthy methods were proposed to offer a panacea to intrusion detection, yet it remains a challenge to build a secure system as the attackers subsequently change their behavior in order to dodge the security mechanism of the system. In current research work, machine learning (ML) classifiers, namely support vector machine (SVM), K-nearest neighbor (KNN), logistic regression (LR), Naïve Bayes (NB), Multi-layer Perceptron (MLP), random forest (RF), Extra-tree classifier (ETC) and decision tree (DT) were employed for classification of data as normal or intrusive. Model performance was analyzed on four feature subsets extracted from the NSLKDD dataset. As the effectiveness of an IDS is determined by dimensions of the data, therefore, data was pre-processing to remove unrelated attributes from the dataset. Empirical results reveal that the performance of RF, ETC, and DT was above 99% for all attack classes using different feature subsets. So, the proposed model has an effective prediction rate, and also reduces the computational complexity by removing irrelevant features.