Beyond the Black Box: Employing LIME and SHAP for Transparent Health Predictions with Machine Learning Models

In the vast realm of healthcare, healthcare data gathered from patients is bountiful. With the continuous evolution and expansion of artificial intelligence, these healthcare data are a vital asset for us. Under the assistance of artificial intelligence, we can efficiently diagnose and prognose diseases to combat the increase in inaccurate prognosis and delayed diagnosis. In healthcare, diagnosis refers to identifying diseases or conditions in patients, while prognosis predicts the likely course and outcome of the medical conditions. To ease the diagnosis and prognosis, we explore the implementation of Machine Learning (ML) techniques and Simple Feedforward Neural Network. The machine learning models that are evaluated include Decision Tree (DT), Random Forest (RF), Support Vector Classifier (SVC), K-Nearest Neighbors (KNN), and Gaussian Naive Bayes (GNB). After evaluation, the KNN model achieved the highest accuracy of 98.56%, along with F1-Score of 98.53%, Precision of 98.69%, and Recall Score of 98.52%. Later, we interpret the decision-making process of the machine learning algorithms by implementing Explainable Artificial Intelligence (XAI). LIME and SHAP, two types of XAI, are employed to explain and visualize the diagnosis capability and feature impact on the models.