Optimal Sampling for Class Balancing with Machine Learning Technique for Intrusion Detection System

Information security is becoming a more important issue in modern computer system. Intrusion Detection System (IDS) as the main security defensive technique that can effectively expand the scope of defense against network intrusion. Data Mining and Machine Learning techniques proved useful and attracted increasing attention in the network intrusion detection research area. Recently, many machine learning methods have also been applied by researchers, to obtain high detection rate. Unfortunately a potential drawback of all those methods is that how to classify attack or intrusion effectively. Looking at such inadequacies, the machine learning technique on balanced classes of data is applied for obtaining the high detection rate. Also, use of internet is increasing progressively, so that large amount of data and it security is also an issue. Sampling technique is one the solution for large datasets. This work proposes a sampling technique for obtaining the sampled data. Sampled dataset represent the whole dataset with proper class balancing. Imbalanced classes can be balanced by sampling techniques. The purpose of this paper is to propose attack classification framework based on a different model. This model also based on machine learning and sampling to improve the classification performance. The proposed work is tested on basis of Accuracy, Error rate, Detection rate and False Alarm rate. KDD CUP'99 dataset used for the approach proposed in this paper. This work suggests the framework for classification of abnormal and normal data and detect intrusions even in large datasets with short training and testing times.