Network Intrusion Detection Using Machine-Learning Techniques

Efficient processing of enormous data is becoming a very serious concern for a real-time Network Intrusion Detection System (IDS) with increasingly large and high-dimensional network traffic. In a real-life scenario, it is very crucial to design an IDS that can prevent intrusions with high accuracy and low-resource usage within the minimum frame of time. This paper evaluates the performance of rule-based and tree-based machine-learning algorithms in order to classify the network traffic as normal or intrusion. Evaluation of these algorithms is performed with the help of NSL-KDD data set in WEKA tool. The objective of this paper is to come up with a novel machine-learning model that achieves high accuracy and takes minimum time in building the model. To achieve this objective, the author aims to select a subset of most significant features and then utilise this subset to determine intrusions in real time with high accuracy. In this paper, feature selection techniques are evaluated with the algorithms on NSL-KDD data set and results are analysed. Experimental results clearly show that when algorithms are evaluated with 10 most important features determined by filter method out of all 41 features, the accuracy of algorithms remains same but model-building time reduced drastically. In fact, in the case of Random Tree when used with a compact set of features suggested by filter method then accuracy is increased up to 99.81% and model-building time is decreased to 0.34 s on the NSL-KDD dataset.