Detecting Network Intrusion by Combining DBSCAN, Principle Component Analysis and Ranker

The internet technology has grown rapidly, which may cause problems in computer network systems. Intrusion Detection System (IDS) has been introduced for detecting attacks in that system. Using artificial intelligence, Intrusion Detection System will able to recognize anomalies or signatures of attacks. For some years, research has focused on the data mining method for detecting the anomaly to increase the accuracy of classification results. To obtain high accuracy, required several stages of data preparation and feature selection. It because the feature selection are not correlated to the dataset and not accordance with requirement of the classification process. In this research, we perform anomaly detection on IDS using combination of DBSCAN, Principle Component Analysis (PCA) and Ranker with classification method. For evaluation, we employ Kyoto 2006, NSL-KDD 99 and KDD Cup 99 as datasets. It is found that this preprocessing step increases the accuracy, when it is applied to Naïve Bayes, Random Forest and k-NN methods. Specifically, the highest increase of accuracy is achieved by Naïve Bayes Classification method with KDD Cup 1999 which gains 6.11%.