Payments Fraud Model
The Challenge: Detecting fraudulent transactions in highly unbalanced datasets (e.g., <1% fraud rate) to minimize financial loss.
The Solution: Built a supervised learning model (XGBoost/Random Forest) utilizing SMOTE for oversampling and cost-sensitive learning.
The Impact: Balanced high Recall (catching fraud) with Precision (minimizing false alerts), resulting in significant projected profit increases and loss reduction.
Patient Utilization & Delivery: Healthcare
The Challenge: Optimizing hospital bed utilization, reducing readmission rates, and managing doctor/nurse burnout.
The Solution: Used predictive analytics on electronic health records (EHR) to identify high-risk surgical patients and workflow bottlenecks.
The Impact: Targeted post-discharge support led to a substantial reduction in readmission rates and improved overall care delivery and organizational performance.
