Detecting insider threat within institutions using CERT dataset and different ML techniques

Abstract

The reason of countries development in industrial and commercial enterprises fields in those countries. The security of a particular country depends on its security institutions, the confidentiality of its employees, their information, the target's information, and information about the forensic evidence for those targets. One of the most important and critical problems in such institutions is the problem of discovering an insider threat that causes loss, damage, or theft the information to hostile or competing parties. This threat is represented by a person who represents one of the employees of the institution, the goal of that person is to steal information or destroy it for the benefit of another institution's desires. The difficulty in detecting this type of threat is due to the difficulty of analyzing the behavior of people within the organization according to their physiological characteristics. In this research, CERT dataset that produced by the University of Carnegie Mellon University has been used in this investigation to detect insider threat. The dataset has been preprocessed. Five effective features were selected to apply three ML techniques Random Forest, Naïve Bayes, and 1 Nearest Neighbor. The results obtained and listed sequentially as 89.75917519%, 91.96650826%, and 94.68205476% with an error rate of 10.24082481%, 8.03349174%, and 5.317945236%.