Anomaly Detection

Health claims Fraud Detection

Project description: Highlight anomalous claims and extract data rules to understand them and help improve the overall claim analysis process.

1. Data

• Batch ETL pipeline to refresh claims features.

2. Model

• Unsupervised Ensemble of multivariate models (Isolation Forest, KNN, UMAP-HDBSCAN…) and univariate statistics (Median Absolute Deviation (MAD), standard deviation…)
• Supervised Decision Tree and RuleFit models on predicted anomalies class to extract and understand the most important rules

Back