Intelligent Insider Threat Detection Using Machine Learning on CERT Dataset

Abstract

This work presents a machine learning-based approach for identifying insider threats using the CERT dataset. The system allows users to upload and process large-scale data that includes multiple behavioral features and class labels. After preparing the dataset, it is divided into training and testing portions to build and evaluate different models. Algorithms such as Random Forest, AdaBoost, XGBoost, LightGBM, and CatBoost are applied to understand their effectiveness. Their performance is measured using accuracy, precision, recall, and related metrics. A comparison graph is used to clearly show differences between models. Among all, CatBoost provides the best results in terms of accuracy. The system also supports prediction on test data, classifying activities as either normal or insider attacks. This approach shows how machine learning can assist in strengthening security systems by identifying unusual behavior patterns.