Performance Evaluation of Tree-Based Classification Techniques on Intrusion Dataset

Abstract

Tree-based learning techniques are extensively used to solve classification problems on various domains. The underlying reason for that is due to its simplicity in implementation, easy to visualize, and more reliable compared to other learning techniques. To select an efficient classification algorithm for predictive analysis on intrusion detection is a challenging task. In this paper, we have tested 13 well-known tree-based classification techniques. The objective is to select the best algorithm among them for intrusion detection model. The tree-based classification techniques are used for prediction of attacks and normal instances on NSL-KDD dataset. The assessment of the methods are evaluated as per confusion matrix, mean absolute error, root mean square error, kappa statistics, ROC curve, accuracy, and execution time. The Weka API’s are used with Java environment to implement the model. Experimentally, the LAD tree methods outperform in comparison with other tree-based algorithms.