cert-dataset

A controlled, reproducible comparison of a hybrid quantum-classical neural network with a parameter-matched classical network for detecting malicious insiders in enterprise activity logs.

Python toolkit for feature extraction, temporal representations, and baseline modeling on CERT insider-threat datasets.

A Proof of Concept data pipeline and Machine Learning model for detecting insider threats. It processes massive CERT corporate logs and uses Markov Chains to model user behavior, calculating log-likelihoods to flag anomalies when activities deviate from historical baselines.

Reproducible zero-shot transfer study from CERT insider-threat logs to LANL authentication data with drift and capacity-constrained model analysis.

Two-stage insider threat detection on CERT r4.2: lightweight XGBoost screening + LLM-based reasoning over graph-informed narratives. University of Ottawa, CSI 5388.

Insider-threat UEBA on CERT r4.2: robust own-history and peer baselines, Isolation Forest and rules, evaluated as an analyst's daily review budget with no look-ahead, and a documented leak in the dataset (the web request count gives the insiders away).