Abstract
While most security projects have focused on fending off attacks coming from outside the organizational boundaries, a real threat has arisen from the people who are inside those perimeter protections. Insider threats have shown their power by hugely affecting national security, financial stability, and the privacy of many thousands of people. What is in the news is the tip of the iceberg, with much more going on under the radar, and some threats never being detected. We propose a hybrid framework based on graphical analysis and anomaly detection approaches, to combat this severe cyber security threat. Our framework analyzes heterogeneous data in isolating possible malicious users hiding behind others. Empirical results reveal this framework to be effective in distinguishing the majority of users who demonstrate typical behavior from the minority of users who show suspicious behavior.
Introduction
The "insider threat" represents a significant cybersecurity challenge: individuals with authorized access—employees, contractors, or partners—misusing their privileges to cause harm. Unlike external attackers, insiders operate behind perimeter defenses, making them "known unknowns." Traditional data-driven security often fails because human behavior is inherently unpredictable. This paper introduces a hybrid framework that shifts from simple log monitoring to a behavior-driven approach, mapping heterogeneous data into multi-layered subgraphs to isolate suspicious outliers.
System Architecture
The proposed framework consists of two primary functional components designed to process massive volumes of noisy log data into actionable security insights:
- Graphical Processing Unit (GPU): This module (a functional unit, not hardware) ingests multidimensional data—such as logon/logoff records and web history—and transforms it into a weighted, undirected bipartite graph. It specifically generates induced subgraphs for every user, analyzing their interaction "neighborhood" up to the 5th order.
- Anomaly Detection Unit (ADU): This unit employs the Isolation Forest (iForest) algorithm. iForest is uniquely suited for this task as it explicitly isolates anomalies rather than attempting to profile "normal" behavior. Its linear time complexity ensures scalability for large enterprise datasets.
Methodology and Data
The study utilized the CERT Insider Threat Dataset (R4.2), encompassing 1,000 employees over 17 months. The framework calculates 50 distinct input parameters across four main categories:
- Graph/Subgraph Properties: Vertex degree, edge count, density, and peer counts.
- Time-Dependent Behavior: Statistical measures (min, max, mean, mode) of logon/logoff and USB usage times.
- Web Access: Volume of unique URL interactions.
- Psychometric Data: Scores based on the "Big Five" personality traits, acknowledging that behavioral precursors often precede technical attacks.
Key Insights and Results
By analyzing 1st through 5th-order subgraphs, the researchers identified subtle clusters of behavior that are invisible when looking at users in isolation. Higher-order subgraphs revealed distinct peaks in activity that helped differentiate typical users from potential threats.
Key Findings:
- Effective Separation: The framework successfully categorized over 79% of users as exhibiting "typical" behavior.
- Anomaly Scoring: Users with an anomaly score (s) above 0.7 were flagged as outliers, with a specific threshold of 0.8 identifying high-risk individuals for investigation.
- Multi-Parameter Evasion: Interestingly, no single malicious user was flagged across more than three parameter categories simultaneously. This confirms that sophisticated insiders often hide their tracks in specific areas, necessitating a multi-dimensional analysis.
Conclusion and Future Work
While effective, the current framework focuses on metadata. Future iterations aim to incorporate deep content analysis of emails and web text, integrate organizational social network data, and transition from static statistical analysis to temporal graphs that track how relationships evolve in real-time. The ultimate goal is a fully automated, scalable system for proactive enterprise monitoring.
Summary updated