Enhancing Business Process Anomaly Detection With Neo4j and Graph Neural Networks
summary
AnomalyNet framework converts BPM event logs into graphs stored in Neo4j graph database and runs graph neural network autoencoders to flag deviations. Experiments use a private tenant dataset plus BPIC 2012 and 2017; a percentile based success rate metric evaluates thresholds. Semi‑supervision recovers all 995 injected subsequence anomalies, word embeddings boost event‑level detection by 20% on BPIC2017, and an edge‑conditioned decoder roughly doubles GCN performance.
AI-written summary · about
chapters (10)
- 0:00Company overview
- 1:44Defining anomaly detection
- 2:44AnomalyNet framework
- 4:07Methodological preliminaries
- 10:12Initial isolation study
- 11:09Semi-supervised GAE research
- 16:09Semantic embedding integration
- 18:39Edge-focused reconstruction
- 22:17Edge-assisted decoder models
- 25:12Explainable AI and closing
related talks
48mHunting Stealth Adversaries with Graphs & AI
26mAdvancing Financial Fraud Detection with Graph Dbs
17mEnhancing Benefit Adjudication Through Graph Node Embedding, Clustering, and Outlier Detection
40mHow to Map Networked Devices Using Graph-Based AI to Preempt Cyber Threats
18mTransforming Financial Decision Intelligence Through Advanced NNs
26mAnomaly Detection System for the CERN Cloud
12mCloud-Native Friendly Fraud Detection
08mAI-Powered Anomaly Detection & Self-Healing Systems