How deeepikashri/credit-risk-ml works
This repository is a command-line credit-default classifier: training generates and splits data, builds and evaluates a scikit-learn pipeline, optionally tunes it, reports metrics and feature importance, and saves the fitted pipeline. Prediction loads that artifact and returns a default probability and risk label for supplied customer features. The README documents the overall workflow and outputs; source excerpts show the training and prediction calls. The tree does not include the documented src/credit_risk layout, so paths below use the actual root-level files.
Main entry points: train.py Training orchestration · predict.py Prediction interface · pipeline.py Preprocessing and forest · config.py Feature configuration
Components 14
Data and configuration
- Feature configuration
config.py - Synthetic data
data.py
Model pipeline
- Preprocessing and forest
pipeline.py - Hyperparameter search
pipeline.py
Training and analysis
- Training orchestration
train.py - Cross-validation
train.py - Evaluation metrics
evaluate.py - Feature importance
explain.py - Metrics and plots
metrics.json
Prediction
- Serialized pipeline
- Prediction interface
predict.py - Risk scoring
predict.py
Other components
- CLI user
- Customer features
Connections 17
- CLI user → Training orchestration: runs training
README.md - Training orchestration → Synthetic data: generates and splits
train.py - Synthetic data → Feature configuration: uses feature definitions
data.py - Training orchestration → Preprocessing and forest: builds and fits
train.py - Preprocessing and forest → Feature configuration: uses feature lists
pipeline.py - Training orchestration → Cross-validation: scores folds
train.py - Training orchestration → Hyperparameter search: optionally tunes
train.py - Hyperparameter search → Preprocessing and forest: searches pipeline parameters
pipeline.py - Training orchestration → Evaluation metrics: computes metrics
train.py - Training orchestration → Feature importance: computes importance
train.py - Training orchestration → Serialized pipeline: serializes fitted model
train.py - Training orchestration → Metrics and plots: writes metrics and plots
train.py - CLI user → Prediction interface: runs prediction
README.md - Customer features → Prediction interface: provides features
README.md - Prediction interface → Serialized pipeline: loads model
predict.py - Prediction interface → Risk scoring: scores customers
predict.py - Risk scoring → Feature configuration: uses features and threshold
predict.py