Diabetes prognosis using white-box machine learning framework for interpretability of results

Artificial intelligence solutions in the healthcare sector are a fundamental phenomenon. It has enabled medical practitioners to perform high quality and precision treatments to prevent diseases or cure a patient. While it is essential to use such solutions, it is also more important to make these solutions transparent to medical professionals. Doctors rely on the cause behind a prognosis rather than just the binary result. This study provides an insight into the feasibility and importance of explainable artificial intelligence solutions for the healthcare sector. A case-study on diabetes in Pima Indian females aids this research motive. The study has maintained good explainability of the predictions and high accuracy by the machine learning models used. This study used a white-box machine learning framework, local interpretable model-agnostic explanations, to prove the cause. The framework successfully interpreted case-by-case predictions of some machine learning models. The machine learning models, while being interpretable, also provided high accuracy in prediction. The highest accuracy, 80.5%, was shown by a random forest model. The study found out glucose levels as the most contributing factors for the outcome of diabetes. The results from this study can be used by researchers to re-evaluate their position on white-box machine-learning solutions in the healthcare sector.