Ablation Study Results for APT Detection via Insider Threat Data

Adel Alshamrani published experimental results on June 2, 2026, for a machine learning approach to detecting Advanced Persistent Threats (APTs). The results, which include a 96.3% detection accuracy and a 42% reduction in false positives, are derived from a methodology that simulates APT behaviors using the CERT Insider Threat Dataset. The dataset is a 5.5 KB Excel file containing ablation study results from a 5-fold cross-validation.

Use Cases

• Benchmarking APT detection models based on the reported 96.3% accuracy and false positive reduction metrics.
• Studying cross-domain learning opportunities between insider threat and APT behaviors as described in the methodology.
• Evaluating the integration of multi-modal data analysis and language models for log analysis in security applications.
• Implementing uncertainty-aware threat detection using techniques like deep evidential clustering mentioned in the research.

Strengths

• Reported high-performance metrics, including 96.3% detection accuracy.
• Demonstrated a 42% reduction in false positives compared to state-of-the-art methods.
• Methodology is based on a known dataset (CERT Insider Threat Dataset) for simulation.
• Results are structured from a 5-fold cross-validation, providing mean and standard deviation metrics.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• The 5.5 KB size indicates a very small, likely summary-level dataset, not the raw training data.

Provenance

Source

Adel Alshamrani via figshare.

Collection Method

Results from an ablation study of a novel machine learning methodology for APT detection.

Freshness

Last updated 2026-06-02 17:27:33; freshness should be verified.