TsoareloAdriaan11/IFRS9-ECL-ENGINE

Next-Generation Expected Credit Loss Engine for South African Banking # 🏦 IFRS 9 Expected Credit Loss (ECL) AI Engine ## 📌 Executive Summary This project is an enterprise-grade, end-to-end quantitative risk data science or credit risk model pipeline designed to calculate Expected Credit Loss (ECL) provisions in strict accordance with the **IFRS 9 Regulatory Standard**. Unlike traditional static actuarial models, this engine integrates **Machine Learning** to dynamically predict Probability of Default (PD) and detect behavioral anomalies, merging advanced data science with strict banking regulations. The project encompasses a full data engineering pipeline, machine learning modeling, an actuarial math engine, automated testing, and an interactive executive web dashboard. ## 🤝 Collaboration This enterprise quantitative risk engine was architected and developed in collaboration with **Masoma Shai**. --- ## 📸 Executive Dashboard & Auditing ### IFRS 9 Portfolio Risk Dashboard *Interactive UI built with Streamlit allowing risk managers to filter the portfolio by IFRS 9 Stage, view total Exposure at Default (EAD), and sort high-risk accounts.* ### SHAP Explainability Auditor Report *AI Model auditing using SHAP values to explain the driving factors behind Probability of Default predictions, ensuring regulatory transparency.* --- ## ⚙️ System Architecture & Pipeline The system is built sequentially across 5 distinct phases: 1. **Synthetic Data Generation (`data_generator`)** * Generates a realistic, synthetic retail/SME banking portfolio. * Creates core variables including FICO equivalents, Days Past Due (DPD) buckets, Loan-to-Income ratios, and outstanding balances. 2. **AI Behavioral Anomaly Detection (`ml_pipeline`)** * Utilizes **Isolation Forests** to scan debtor behavioral data for hidden risk anomalies prior to outright default. 3. **AI Probability of Default Modeling (`ml_pipeline`)** * Utilizes an ML classifier to calculate a granular, forward-looking Probability of Default (PD) for every individual loan. 4. **IFR …