Benchmark Datasets for Network Intrusion Detection: A Review

Network Intrusion Detection is the process of monitoring the events occurring in a computer system or the network and analyzing them for the signs of possible intrusions. An intrusion is a potentially harmful activity of malicious user, aimed at compromising the confidentiality, availability and integrity of the system. Over the decades intrusion detection (ID) problem has been visited by the researchers in various available environments like finite state automata, rule based systems, Markov probabilis-tic approach, statically sought solutions and most popular of all data mining and machine learning techniques. The prerequisite for data mining is that data should be present and there should be some hidden patterns in the data which need to be unearthed. In this work, we intend to provide a thorough review of the benchmark datasets available for Network Intrusion Detection (NID) which researchers in the field can use to train and test their models. In addition, this work as the first of its kind implements k-NN a simple most instance based type of classifier over all the datasets that doesn't require a well planed and monolithic training phase, across different neighborhood sizes. Results show that off all the datasets k-NN performs better on NSL-KDD dataset due to the fact that NSL-KDD doesn't have any redundant network connections and connections being fairly distributed across all the classes.