Interpretable Machine Learning Models for Predicting Malaria

Malaria remains one of the deadliest diseases in underdeveloped regions, particularly in Sub-Saharan Africa. The lack of high-quality healthcare services and accurate disease diagnosis systems has resulted in acute medical problems for patients. This necessitates reliable automated decision-making tools to aid medical professionals in their decision-making process. This paper presents a transparent approach to malaria diagnosis by applying Explainable Artificial Intelligence (XAI) techniques, namely Shapley Additive Explanation (SHAP) and Local Interpretable Model-agnostic Explanation (LIME), to provide meaningful interpretations of severe malaria predictions made by machine learning models. Various models, including Extreme Gradient Boosting, K-means, K-Nearest Neighbor, Support Vector Machine (SVM), Decision Tree, Logistic Regression (LR), Random Forest, Naive Bayes, AdaBoost, and Explainable Boosting Machines (EBMs) are deployed for this task. The results of the study showed that Random Forest and Explainable Boosting Machines achieved the highest accuracy of 84%. EBM also provided a practical clinical understanding of features that drive clear prediction. The LR achieved an accuracy of 81% after applying GridSearchCV to increase prediction accuracy. Furthermore, K-fold validation was used on XGBoost to estimate the model’s skill on new data. The interpretations were enhanced by XAI, which revealed features that contribute to severe malaria. The application of these techniques can significantly improve the accuracy of severe malaria predictions and aid medical professionals in making informed decisions. This paper provides a compelling argument for the urgent need for XAI techniques to address the challenges associated with severe malaria diagnosis and treatment. The study’s findings demonstrate the effectiveness of these techniques in enhancing the accuracy and interpretability of machine learning models, which can greatly benefit medical professionals in their decision-making process.