A Parallel and Scalable Framework for Insider Threat Detection

Abstract

In this article, we propose an innovative method for the detection of insider threats. This method is based on a unite and conquer approach used to combine ensemble learning techniques, which have the particularity of being intrinsically parallel. Furthermore, it showcases multi-level parallelism properties, offers fault tolerance, and is suitable for heterogeneous architectures. To highlight our approach's efficacy, we present a use case of insider threat detection on a parallel platform. This experiment's results showed the benefits of this method relative to its improvement of classification AUC-score and its scalability.