Zoish

Zoish applies Shapley additive values to perform feature selection for healthcare analytics, providing interpretable and bias-aware feature attributions for predictive modeling.

Key Features:

• Shapley Additive Values Integration: Implements Shapley additive values from cooperative game theory to quantify each feature's contribution to model predictions.
• Versatile Compatibility: Compatible with scikit-learn, XGBoost, CatBoost, and imbalanced-learn for integration with common machine learning workflows.
• Dual Algorithmic Approach: Implements a dual algorithmic strategy for calculating Shapley values to handle both large and small datasets efficiently.
• Interpretability and Visualization: Generates visualizations that represent feature attributions and support interpretability of models.
• Customizable Settings: Offers configurable parameters to control feature selection criteria and optimize for specific predictive objectives.

Scientific Applications:

• Breast cancer prediction: Applied to breast cancer prediction case studies to assess predictive feature selection.
• MoCA prediction in Parkinson's disease: Applied to Montreal Cognitive Assessment (MoCA) prediction in Parkinson's disease to identify predictive features.
• Synthetic dataset evaluation: Evaluated on 300 synthetic datasets to assess robustness and adaptability across varied data conditions.

Methodology:

Computationally, Zoish employs Shapley additive values, uses a dual algorithmic strategy for Shapley value calculation, and combines local and global feature selection into a single process.