IterSHAP: Iterative feature selection using SHAP values

Author: Frank van Mourik, University of Twente

Installation

Install via pip using: pip install itershap (requires Python version >=3.10,<3.11).

Usage

from itershap import IterSHAP

X, y = get_data() # Replace with data location

fs = IterSHAP() # Create a IterSHAP feature selection object
fs.fit(X, y) # Execute IterSHAP on input data
X_transformed = fs.transform(X) # Only keep the via IterSHAP selected features

Benefits

• Performs well on small high-dimensional datasets
• Guarantees to return a feature subset
• Model-agnostic (limited by shap supported models)
• Validated on synthesised data
• Benchmarked on DEAP dataset

License

Available under the MIT license, which can be found here