ML Regression Benchmark

A compact, reproducible baseline for comparing regression models on a clearly labelled synthetic dataset.

What it does

• Creates a synthetic regression dataset with a fixed random seed.
• Uses a train/test split to keep final evaluation data separate from model fitting.
• Compares Dummy Regressor, Linear Regression, Random Forest, and HistGradientBoosting.
• Reports MAE, RMSE, and R² in a CSV file.
• Saves a model-comparison plot and prints the best model by test RMSE.

This is a demonstration project, not a claim of performance on a real-world or NASA battery dataset.

Requirements

Python 3.10+ recommended.

python -m pip install -r requirements.txt

Run

From the project root:

python -m src.benchmark

Outputs are written to outputs/:

• regression_metrics.csv
• regression_benchmark.png

Tests

python -m unittest discover -s tests -v

Reproducibility and limitations

The dataset is synthetic and generated with a fixed seed. The test split is held out from model fitting. The benchmark is intended as a small baseline template; real projects should use domain-appropriate validation, leakage checks, and uncertainty analysis.

License

MIT. See LICENSE.