CPA Finalist building credit-risk and fraud models for development-finance lenders.
| Project | What it does | Result |
|---|---|---|
| Credit Risk Scorecard (live) | From-scratch WoE/IV and logistic regression checked against scikit-learn and scipy at every step (99.9997% prediction agreement), on 307,511 real Home Credit applicants plus bureau/previous-application history | 0.762 AUC, 0.394 KS |
| Fraud Detection System (live · dashboard) | XGBoost fraud classifier on 6.3M PaySim mobile-money transactions: balance-discrepancy feature engineering, isotonic calibration, walk-forward validated across 4 folds | 99.85% precision / 99.56% recall |
| Kiva Loans Microfinance Analytics | Funding-risk model on 671K real Kiva microloans joined to region-level MPI poverty data, with SHAP attribution and a days-to-fund regression | 0.4889 PR-AUC, 7.43-day MAE |
| Stock-Portfolio-Tracker-Analytics-Engine | Portfolio risk/performance analytics engine in Excel: VaR/CVaR, CAPM, Black-Litterman optimisation, tax-aware rebalancing | 12.59% 7-yr CAGR, 0.37 Sharpe, -12.33% max drawdown (23-test validated) |
Python XGBoost scikit-learn SHAP Pandas SQL Excel Power BI
Credit risk work covers WoE/IV, scorecard development, and GINI/KS/PSI validation against IFRS 9 ECL requirements. On the fraud side: imbalanced classification with cost-sensitive thresholding, evaluated PR-AUC-first. Finance modelling spans GAAP/IFRS, 3-statement builds, and DCF valuation.
Open to Credit Risk Analyst, Data Analyst, and Financial Data Scientist roles.
📫 alvenyuka2@gmail.com · 💼 LinkedIn