sitahlango-maker/Financial_Inclusion

Analyzing impact of Micro and Macro factors affecting Financial inclusion in East Africa # 🌍 Digital Finance Access Predictor ## Overview This project investigates whether digital financial inclusion can be predicted using historical socioeconomic and demographic data from East Africa. The study compares multiple machine learning approaches to determine whether specialized country-specific models outperform traditional pooled models. The system predicts the probability that an individual has access to a digital financial account and provides explainable insights into the factors influencing the prediction. ## Research Objective The study seeks to answer the following questions: 1. Can digital financial inclusion be predicted using historical data? 2. Does a country-specific Mixture of Experts (MoE) architecture outperform a traditional pooled model? 3. Does harmonizing country data improve predictive performance? 4. Which socioeconomic factors contribute most to digital financial inclusion? ## Models Evaluated ### 1. Pooled Model A single XGBoost model trained on data from all countries combined. ### 2. Harmonized Model An XGBoost model trained on a balanced dataset where each country contributes an equal number of observations. ### 3. Expert Models Country-specific models trained separately for: * Kenya (KEN) * Tanzania (TZA) * Uganda (UGA) ### 4. Routing Model A Random Forest classifier that dynamically selects the most appropriate model for each observation. ### 5. Mixture of Experts (MoE) The final architecture combining: Input → Router → Best Model → Prediction ## Repository Structure ```text Financial_Inclusion/ ├── app.py ├── routing.py ├── requirements.txt ├── runtime.txt ├── README.md ├── feature_columns.joblib ├── pooled_model.joblib ├── harmonized_model.joblib ├── expert_model_KEN.joblib ├── expert_model_TZA.joblib ├── expert_model_UGA.joblib ├── routing_model.joblib ├── final_model_comparison.csv ├── feature_impact_table.csv ├── final_model_comparison_chart.png ├── feature_impact_comparison.png ├── shap_feat …