Transaction Fraud Detection Workbench for Retail Security Teams

Retail fraud has industrialized through stolen cards, refund abuse, and account takeovers, and mid-size merchants are squeezed between fraud losses and false positives that anger legitimate customers. Fraud teams need detection that learns patterns from their own transaction data and tooling that supports investigation, not just a stream of alerts. The intern builds a fraud detection workbench in Python. Uploaded transaction logs are processed into features, scikit-learn models are trained to score transactions for fraud risk, and detected patterns render as visual risk views in a Vue.js frontend: clusters of suspicious cards, velocity anomalies, and unusual geographic spreads. Real-time alerts surface high-risk transactions as they arrive, and collaborative investigation workflows let analysts assign cases, annotate findings, and record outcomes that feed back into model retraining. MongoDB stores transactions and case data with careful access controls, and Docker packages the suite for deployment. The intern learns the applied machine learning security loop end to end: feature engineering for adversarial domains, evaluating models on heavily imbalanced classes, and building the investigation experience that turns model scores into stopped fraud.