Investigation on effective solutions against insider attacks (B)

Abstract

Insider Attacks (IAs) can be defined as an attack or intrusion that is performed from the internal boundaries of the network. While Intrusion Detection Systems (IDS) devices are placed strategically in the network to detect external intrusions, the same cannot be said for internal intrusions like IAs. Furthermore, existing IDS technologies have proven to be lacking in detecting obscure attacks like IAs. This project investigates the implementation of several ML-based classification algorithms into the proposed IDS, specifically, Support Vector Machine (SVM), Extreme Learning Machine (ELM) and Multi-layer Perceptron (MLP). The performance of the classifiers were analyzed and compared against one another as a means to find the best ML classifier for IDS against IAs and EAs. A Python-based program was created for this project to verify the IDS detection and classification performance of EAs and IAs using two modified datasets, IAs Dataset and EA Dataset, deriving from NSL-KDD dataset. Five experimental trials were conducted, and it was discovered that the ML classifiers exemplified robust performance, with MLP yielding the most effective detection performance, and SVM and ELM yielding strong efficiency performance.