← Back to projects

Machine Learning · Big Data · 2026

Healthcare Fraud Detection System

Real-time ML inference, Medallion data pipeline and an explainable enterprise dashboard.

PythonXGBoostSHAPFastAPIPandasScikit-learnSQL

92.5%

Model accuracy

94.8%

ROC AUC

44+

API endpoints

27

Dashboard pages

The problem

Healthcare insurers lose large amounts to fraudulent claims that are hard to spot manually. Reviewers need not only a risk score, but a clear reason why a claim looks suspicious.

The solution

A production-grade platform that scores every claim with an XGBoost model trained on 20K+ synthetic claims, explains each decision with SHAP feature contributions, and surfaces everything in a role-based dashboard for insurers and providers.

Medallion data architecture

Data flows through three curated layers so every metric is reproducible.

  • Bronze — raw synthetic health claims ingested as-is
  • Silver — cleaned, deduplicated and validated records
  • Gold — aggregated business metrics: claims per patient, average cost, fraud cases

Machine learning

  • XGBoost classifier with 24 engineered features
  • Real-time scoring on claim submission
  • SHAP explanations per claim with severity levels and confidence scores
  • Model management: metrics, version history and retraining triggers

Dashboard experience

  • 22 insurance pages: executive KPIs, analytics, flagged claims, fraud heatmap, audit logs
  • 3 provider pages: submit claims with instant ML scoring and track statuses
  • Alert center, notifications, reports with CSV export and system monitoring

Results & impact

  • 92.5% accuracy and 94.8% ROC AUC on held-out claims
  • Every flag is explainable, so reviewers can justify a decision
  • Geographic and provider-level views reveal fraud clusters, not just single claims

More projects