Assessfy Pvt. Ltd Advanced 5 milestones 100 marks

Healthcare Claims Fraud Detection using Graph Neural Networks

Target year: BE Sem 7-8 (Major Project Phase-I/II) AICTE: 6 credits · ~150 hrs Bloom: Create / Evaluate MU CBCS: DS801 / DSDLO8021 BE Project

Real-world project · AICTE-aligned · AI-graded · Audit-ready certificate

Enroll as Student Mentor this project

Milestones

Available mentors

Enrolled students

Core skills

About this project

Detect fraud in health-insurance claims (overbilling, phantom services, kickback rings) using Graph Neural Networks. Build a heterogeneous graph (patient ↔ provider ↔ claim ↔ procedure code), train GraphSAGE / GAT, compare with tabular XGBoost baseline. Demonstrate +10% recall@5% FPR.

Course Learning Outcomes (CLOs):
CLO1: Apply GNN architectures to a heterogeneous-graph problem.
CLO2: Compare graph-based vs tabular models on the same task.
CLO3: Evaluate with cost-sensitive metrics (recall@k%FPR, lift@top-1%).
CLO4: Engineer features that combine relational + tabular signals.
CLO5: Communicate findings to a domain (insurance) audience.

Industry/societal relevance: Indian health insurance (PMJAY, private insurers like Star Health, HDFC ERGO) processes 100M+ claims annually; fraud-detection is a major hiring area.

Milestones

1. Dataset Construction

15 marks 14d

Use Medicare PUF data (or synthetic Ayushman-style). Construct heterogeneous graph. Validate stats.

2. XGBoost Tabular Baseline

20 marks 18d

Engineer 30+ features per claim. Train XGBoost, report ROC-AUC + recall at 5%-FPR.

3. GraphSAGE on Bipartite

25 marks 21d

Bipartite patient-provider graph. Train GraphSAGE node-level classifier. Compare to baseline.

4. GAT on Heterogeneous Graph

20 marks 21d

Add procedure-code + claim nodes. Apply heterogeneous attention. Beat GraphSAGE on recall.

5. Cost-Sensitive Eval + IEEE Paper

20 marks 26d

At a chosen FPR budget, compute INR savings. 14-page IEEE paper + oral defense.

Open internships using this project -->

Skills you'll learn

Graph Neural Networks (GraphSAGEGAT)PyTorch GeometricAnomaly detectionHeterogeneous graph dataPrivacy + HIPAA-like ethicsCost-sensitive evaluation

Tools used

Python 3.11PyTorch 2.x + PyTorch GeometricNetworkXscikit-learnXGBoostJupyterGitHubGPU access

Prerequisites

Deep Learning; Graph theory basics; Python advanced; intro to anomaly detection

Available mentors

No mentors have signed up for this project yet.

Be the first to mentor

You'll earn — Certificate (PDF)

AICTE-aligned Project Completion Certificate

A formal, audit-ready PDF certificate issued by Assessfy + your institute on successful completion. Includes AICTE credit hours, your evaluator's signature, and a QR code for third-party verification.

AICTE-aligned

Certificate of Project Completion

This is to certify that

has successfully completed the project

Healthcare Claims Fraud Detection using Graph Neural Networ…

Scan to Verify

Date

Signature

Auto-issued on completion QR-verifiable

You'll earn — Digital Badge

Shareable LinkedIn / Resume Skill Badge

A compact, verifiable Open-Badges-2.0-compliant digital credential. Add to your LinkedIn profile, GitHub README, or resume in one click. Recruiters can validate authenticity via a unique URL.

Advanced

Healthcare Claims Fraud Detection usi…

Assessfy

Auto-issued on completion One-click LinkedIn add