Transparent Models in Healthcare: Enhancing Decision Support through Explainability

This paper introduces a framework employing inherently explainable Machine Learning models to provide intuitive global explanations for medical diagnostics. Our proposal uses a transparent approach based on \emph{Soft Decision Trees} that offers direct insights into its decision-making processes, eliminating the need for post-hoc explanation methods. SDTs combine the hierarchical structure of decision trees with the representational power of neural networks, allowing them to capture complex data patterns while maintaining interpretability. In the context of healthcare analytics, we apply SDTs to a classification task. The model is trained end-to-end using mini-batch gradient descent to minimize a cross-entropy loss, encouraging balanced hierarchical data partitioning. An ANOVA-based feature selection technique is implemented to reduce model complexity and enhance interpretability. We tested our methodology on a dataset from a healthcare scenario, demonstrating that SDTs effectively provide precise and understandable diagnostic predictions. The findings emphasize the role of explainable AI in enhancing trust and cooperation in healthcare decision-making.