Experimental analysis of data mining application for intrusion detection with feature reduction

As tremendous growth of information in the internet, the importance of Network security also dramatically increases. Network and Host based Intrusion Detection System (IDS) are two primary systems in Network Security infrastructure. When new intrusion types are appeared in Network or Host, some serious problems are also appeared to detect these new intrusions. Due to this reason, IDSs demanded better than Signature based detection. The action of intrusion is represented by some features and collects the corresponding featured data from these uncertain feature characteristics. In last two decades, several techniques are developed to detect intrusion by using these data as human labeling which is very time consuming and expensive process. In this paper, we proposed a data mining rule based algorithm called Decision Table (DT) to detect intrusion and a new feature selection process to remove irrelevant/correlated features simultaneously. An empirical analysis on KDD'99 cup dataset was performed by using our proposed and some other existence feature selection techniques with DT and some others classification algorithms. The experimental results showed that proposed approach provides better performance in accuracy and cost compared among Bayesian Network, Naïve Bayes Classifier and other developed algorithms with data mining KDD'99 cup challenge in all cases.