At Oasis Infobyte, I built a credit card fraud detection model for highly imbalanced data, reaching 0.802 PR-AUC and 0.926 ROC-AUC on unseen test data. I also cleaned retail transaction records and used RFM features to segment customers.
For my final-year project at AFIT, I built an explainable stacking ensemble that recommends 22 crop types, achieving 99.55% test accuracy. I used cross-model SHAP analysis to identify the main drivers of its recommendations.

