Intrusion Detection Model using PCA and Ensemble of Classiers

Abstract

Most of the intrusion detection systems examine all network features to identify intrusions with different classification approaches. The major challenges for any intrusion detection model is to achieve maximum accuracy with minimal false alarms. While many ensemble techniques are present to improve the accuracy of intrusion detection models, building an ensemble that can be generically applied for any network traffic is still a difficult task. In this paper, we propose a hybrid model for intrusion detection integrating base classifiers such as SVM, Linear Discriminant and Quadratic discriminant analysis. The aim of this paper is to identify the class label by constructing an individual classifier for each of the attack type and merging the results of every classifier. The resultant decision of the class label is obtained using weighted majority voting approach. We analyzed the performance of the model on two different data sets such as NSL-KDD and UNSW-NB datasets. The experimental results indicate that the ensemble produces high accuracy in comparison to the base classifiers. As there is a huge class imbalance problem in network traffic, it is also observed that rather than relying on a single classifier, predicting the class label by weighted majority voting of SVM, Linear and Quadratic Discriminant classifier is an optimal solution which is proposed in this paper.