Farman24476/Competition-Project-African-Credit-Scoring-Challenge-Predicting-Loan-Default-with-Deep-Learning

# Competition-Project-African-Credit-Scoring-Challenge-Predicting-Loan-Default-with-Deep-Learning Verified on Zindi This repository presents a complete machine learning pipeline to predict loan default using customer financial data in the context of an African credit scoring challenge. The goal is to build a binary classifier that can predict whether a customer is likely to default on a loan, with a strong emphasis on precision and F1-score. ## Project Objectives - Build a robust model to predict credit risk (default vs. non-default) - Engineer high-quality interaction features from raw data - Apply deep learning techniques on structured tabular data - Tune the prediction threshold for optimal F1 performance ## Dataset - **Train.csv** – labeled data with customer features and target variable - **Test.csv** – unlabeled data to be used for prediction - **Target** – binary indicator (`1` = will default, `0` = will not default) ## Workflow Overview ### 1. Data Preprocessing - Loaded and inspected `Train.csv` and `Test.csv` - Verified data types and absence of missing values - Separated features and target labels ### 2. Feature Engineering - Generated interaction features using `PolynomialFeatures` (degree 2, interaction-only) - Selected top 30 most important features using `SelectKBest` with ANOVA F-statistic ### 3. Handling Class Imbalance - Calculated class weights to address imbalance in the training labels - Used `compute_class_weight` to feed into model training ### 4. Model Architecture (Deep Neural Network) Constructed a deep feedforward neural network using `TensorFlow/Keras`: - Input layer with 30 selected features - Dense layers: [1024, 512, 256, 128, 64, 128] - Regularization: Batch Normalization, Dropout, L2 penalties - Output layer: Sigmoid activation for binary classification - Optimizer: Adam (learning rate = 0.001) - Loss function: Binary Crossentropy ### 5. Threshold Optimization - Generated predicted probabilities on the validation set - Tu …