georgekusche/credit_african

African Credit modeling challenge out of curiosity and using other solutions as reference. African Credit Score Challenge The objective was to build a model that predicts whether a loan applicant in an African financial dataset is likely to default. Project Contents - notebooks/eda.ipynb: The main notebook that includes all stages of the project – from data loading and cleaning to feature engineering, modeling, evaluation, and prediction. - data/: Contains the provided datasets including train.csv, test.csv, and economic_indicators.csv. - submission.csv: Final output file containing predictions. - requirements.txt: List of all Python dependencies required to run the project. - README.md: This file. How to Run the Project 1. Set up a new Python environment using Conda or virtualenv (recommended). Example using Conda: conda create -n miway_env python=3.10 conda activate miway_env 2. Install all dependencies: pip install -r requirements.txt 3. Launch Jupyter Notebook: jupyter notebook 4. Open the file: notebooks/eda.ipynb 5. Run all cells in the notebook in order. This will: - Load and filter the datasets - Apply feature engineering - Tune models using Optuna - Evaluate the best model using F1-score, precision, recall, and accuracy - Generate predictions for the test set - Export a final submission.csv file containing ID, target, and credit_score Submission File The submission.csv file will contain: - ID: Loan ID from the test set - target: 0 or 1 (predicted default) - credit_score: Predicted probability of default Notes - All code is kept inside one notebook for simplicity. - No external scripts or models are required to reproduce results. - The models tested include LightGBM, XGBoost, and CatBoost. Optuna selects the best model and hyperparameters automatically. - Business rules and domain logic were applied to filter the training and test sets to ensure consistency and improve performance. Author George Kusche