Sg254/Kenya-Energy-AI-System-

# Kenya Energy Access AI System - Documentation ## Model Card: Credit Scoring Model ### Model Details - **Model Name**: Energy Access Credit Scoring Model v1.0 - **Model Type**: XGBoost Gradient Boosted Trees - **Version**: 1.0.0 - **Date**: January 2025 - **Owner**: Kenya Energy AI Team - **Contact**: ml-team@energy-ai.ke ### Intended Use **Primary Use Cases:** - Assess creditworthiness of PAYG (Pay-As-You-Go) energy customers - Optimize payment collection strategies - Identify customers at risk of default - Support expansion of energy access to underserved areas **Target Users:** - Energy providers (utilities, mini-grids, solar companies) - Government policy makers (REREC, KPLC) - Financial institutions - Development organizations **Out-of-Scope Uses:** - Should NOT be used as sole determinant for service denial - Not designed for credit decisions outside energy access context - Not intended for individual loan approval decisions ### Training Data **Data Sources:** - Smart meter readings (60%) - M-PESA payment transactions (25%) - Customer demographic data (10%) - Grid reliability data (5%) **Data Size:** - Training: 500,000 customer records - Validation: 100,000 customer records - Time period: 2023-2024 (24 months) **Data Preprocessing:** - PII anonymization using SHA-256 hashing - Location generalization to 2 decimal places - Missing value imputation using median/mode - Outlier capping at 99th percentile **Feature Engineering:** - Payment behavior features (15 features) - Energy usage patterns (12 features) - Demographic indicators (8 features) - Temporal features (5 features) - **Total Features**: 40 ### Model Architecture ``` XGBoost Parameters: - max_depth: 6 - learning_rate: 0.1 - n_estimators: 100 - min_child_weight: 3 - gamma: 0.1 - subsample: 0.8 - colsample_bytree: 0.8 - objective: binary:logistic - eval_metric: auc ``` ### Performance Metrics **Overall Performance:** - AUC-ROC: 0.82 - Precision: 0.76 - Recall: 0.79 - F1-Score: 0.77 - Accurac …