Efficient Dimensionality Reduction Using Best First Search and Data Mining Classifiers for Intrusion Detection System

The area of intrusion detection is central to the concept of computer security. One major challenge in intrusion detection is that we have to identify the camouflaged intrusions from a huge amount of normal communication activities. It is demands to applying data mining techniques to detect various intrusions. The Data Mining process requires high computational cost when dealing with large data sets. Reducing dimensionality can effectively cut this cost. Hence dimensionality reduction is vital when data mining techniques are applied for intrusion detection. In this paper dimensionality reduction is performed to reduce 41 attributes to 7 attributes based on Best First Search method and the classification of normal and abnormal packets are performed using J48, Id3, Random tree and Naïve Bayes updateable classifiers. These models are tested on KDD Cup 99 dataset. The result shows that J48 classifier outperforms ID3, Random tree and Naïve Bayes Updateable. The experimental result renders remarkable improvement in reducing the false alarms and it also reduces the time complexity.