Project

Fraud / Under-Appraisal Clustering

Machine learning / anomaly detection • Completed

Summary

A clustering analysis to identify suspicious appraisal behavior and potential fraud in financial datasets.

Problem

Financial institutions struggle to detect appraisal outliers quickly without overfitting rules to one set of cases.

Approach

I used dimensionality reduction and unsupervised clustering to group appraisals by feature patterns and then flagged unusual clusters for review.

Results

The project reduced the initial flagged population by 30% while preserving high-risk cases, making analyst review more efficient.