OrkhanIsmayilov992/Africa-Economic-Banking-and-Systemic-Crisis

Machine learning models (Logistic Regression, Decision Tree, Random Forest, XGBoost) to predict systemic banking and economic crises in African countries using historical macroeconomic indicators (1860–2014). # 🌍 Africa Economic, Banking & Systemic Crisis Prediction A machine learning project to predict **systemic economic crises** in African countries. The project analyzes macroeconomic indicators (exchange rate, inflation, sovereign debt defaults, etc.) for 13 African countries between 1860–2014 and compares four different classification models to assess crisis risk. > 🚀 **Open in Google Colab:** Launch Notebook in Colab > 📊 **Dataset:** African Crises Dataset (Kaggle) --- ## 📌 About the project This notebook covers the following steps: 1. **Exploratory Data Analysis (EDA)** — checking for missing values, unique values, and variable distributions 2. **Data cleaning** — converting the `banking_crisis` column from text values (`crisis` / `no_crisis`) into 0/1 format 3. **Visualization** - Frequency plots for crisis types (systemic, currency, inflation, banking crises) - GDP-weighted default indicators by country - Exchange rate and inflation (CPI) trend charts for each country - Comparison of crisis frequency by independence status 4. **Feature selection and data split** - Target variable: `systemic_crisis` (more balanced class distribution — ~8-10% positive) - Features: `exch_usd`, `inflation_annual_cpi`, `gdp_weighted_default` - **Chronological split**: pre-1999 for training, 1999+ for testing (to avoid data leakage) 5. **Handling class imbalance** — SMOTE (Synthetic Minority Over-sampling) 6. **Model training** — four different models compared: - Logistic Regression - Decision Tree - Random Forest - XGBoost 7. **Evaluation** — comparing model performance using Accuracy, Precision, Recall, and F1-score --- ## 📈 Results | Model | Recall (Class 1) | Notes | |---|---|---| | Logistic Regression | ~0.90 | High recall, but very low precision | | Decision Tree | ~0.15 (Best F1: 0.169) | Most balanced model | | Random Forest | ~0.25 | Good overall performance | | XGBoost | ~0.21 | Captures non-linear patterns well | **Recommended model:** Decision Tree — provides t …