An Explainable Machine Learning Framework for Insider Threat Detection Using Behavioral Feature Engineering

Insider threat detection remains one of the most challenging problems in cybersecurity because malicious activities are often difficult to distinguish from legitimate user behavior. Many existing machine learning approaches rely on limited behavioral representations, high-dimensional feature spaces, and black-box models with limited interpretability, which constrains their practical utility in security operations. This paper proposes an explainable machine learning framework for insider threat detection based on behavioral feature engineering. The framework constructs daily user-level behavioral profiles from heterogeneous sources in the Computer Emergency Response Team (CERT) Insider Threat Dataset (Release 4.2), including logon, email, web browsing, file access, Universal Serial Bus (USB) device usage, and psychometric records. A Genetic Algorithm (GA) reduces the engineered feature space from 53 to 35 features while retaining comparable predictive performance. A LightGBM classifier is then optimized through Bayesian hyperparameter tuning with Optuna, followed by decision threshold optimization to balance precision and recall. SHapley Additive exPlanations (SHAP) provide global and local explanations, enabling security analysts to examine the factors behind individual decisions. On the independent test set, the framework achieved 99.79% accuracy, 83.78% precision, 60.78% recall, 70.45% F1-score, 97.92% ROC-AUC, 71.14% PR-AUC, 80.37% balanced accuracy, and 71.27% MCC. The gap between high accuracy and lower recall and F1-score reflects the severe class imbalance of the dataset. These findings demonstrate the feasibility of integrating behavioral feature engineering, feature selection, model optimization, and explainable AI within a unified framework.

Authors

Institutions

Publication Details

Journal
Dasinya Journal for Engineering and Informatics
Published
2026-11-13
DOI
https://doi.org/10.65542/djei.v2i4.102
Primary Topic
Network Security and Intrusion Detection
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

An Explainable Machine Learning Framework for Insider Threat Detection Using Behavioral Feature Engineering

Omar Sedqi Kareem
3 links
Dasinya Journal for Engineering and Informatics
Network Security and Intrusion Detection
article

An Explainable Machine Learning Framework for Insider Threat Detection Using Behavioral Feature Engineering

Omar Sedqi Kareem
article en

Abstract

Insider threat detection remains one of the most challenging problems in cybersecurity because malicious activities are often difficult to distinguish from legitimate user behavior. Many existing machine learning approaches rely on limited behavioral representations, high-dimensional feature spaces, and black-box models with limited interpretability, which constrains their practical utility in security operations. This paper proposes an explainable machine learning framework for insider threat detection based on behavioral feature engineering. The framework constructs daily user-level behavioral profiles from heterogeneous sources in the Computer Emergency Response Team (CERT) Insider Threat Dataset (Release 4.2), including logon, email, web browsing, file access, Universal Serial Bus (USB) device usage, and psychometric records. A Genetic Algorithm (GA) reduces the engineered feature space from 53 to 35 features while retaining comparable predictive performance. A LightGBM classifier is then optimized through Bayesian hyperparameter tuning with Optuna, followed by decision threshold optimization to balance precision and recall. SHapley Additive exPlanations (SHAP) provide global and local explanations, enabling security analysts to examine the factors behind individual decisions. On the independent test set, the framework achieved 99.79% accuracy, 83.78% precision, 60.78% recall, 70.45% F1-score, 97.92% ROC-AUC, 71.14% PR-AUC, 80.37% balanced accuracy, and 71.27% MCC. The gap between high accuracy and lower recall and F1-score reflects the severe class imbalance of the dataset. These findings demonstrate the feasibility of integrating behavioral feature engineering, feature selection, model optimization, and explainable AI within a unified framework.

Dasinya Journal for Engineering and InformaticsVol. 2(4)
Duhok Polytechnic University (IQ)
Network Security and Intrusion Detection
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon