An Explainable Machine Learning Framework for Multiple Medical Datasets Classification
Mostarina Mitu, S. M. Mahedy Hasan, Anwar Hossain Efat, Md Fakrul Taraque, Nahrin Jannat, Mahjabin Oishe
Abstract
Machine learning (ML) has emerged as a ground-breaking approach for disease prognostication, garnering considerable attention from researchers in recent times. Although separate disease-specific models have been developed, but it is very challenging to analyze the predictions of black-box ML classifiers to identify the most prominent risk factors and build a generalized model that can effectively detect commonly occurring diseases (i.e., Diabetes, Breast Cancer, and Heart Disease) with a high recognition rate. In this paper, we use a ML based approach for detecting these diseases and the Shapley Additive Explanations Framework (SHAP) for analyzing the black-box ML classifiers. We use six traditional ML classifiers and develop a weighted averaging (WA) ensemble model for this purpose. An empirical evaluation based on three publicly available datasets demonstrates that our proposed approach detects these three diseases with high accuracy, precision, and recall. The results obtained from this study can be valuable in creating more effective methods for diagnosing patients at an early stage and preventing the progression of diseases, thereby improving treatment outcomes and patient well-being.