$ cat projects/creditsense-ai/README.md
CreditSense AI
A leak-free ML pipeline that scores loan eligibility — benchmarking five classifiers with stratified cross-validation to find the most reliable approve / reject model.
$ less about.txt
CreditSense AI turns a raw loan dataset into a rigorous, reproducible model comparison. A scikit-learn Pipeline and ColumnTransformer do all the preparation inside cross-validation — dropping identifiers, imputing missing values (median for numbers, most-frequent for categories) and one-hot encoding — so no information ever leaks from validation folds into training. Five classifiers (KNN, Logistic Regression, Decision Tree, an MLP neural network and Random Forest) are each scored with 5-fold stratified cross-validation and out-of-fold predictions, using adaptive scaling (MinMax for KNN, standardisation for the rest). Every model is judged on accuracy, precision, recall, F1 and ROC-AUC against a majority-class baseline (~0.69), with confusion matrices and ROC curves saved as figures. A command-line project built to do the evaluation honestly rather than chase a single lucky train/test split.
$ open architecture.webp
