Enhancing Monkeypox Diagnosis: Explainable AI-Driven Evaluation of Machine Learning Models for Reliable and Transparent Detection

Abstract

Monkeypox is a re-emerging zoonotic disease caused by the monkeypox virus (MPXV), which raises serious public health issues owing to the possibility of human-to-human transmission. Traditional diagnostic methods, such as PCR and serological testing, are effective but time-consuming and resource-intensive, limiting rapid outbreak response. Machine learning-based systems are a viable alternative, but their lack of interpretability and transparency remains a significant limitation in clinical decision-making. This study addresses the challenges by applying four machine learning algorithms: AdaBoost, Gradient Boosting Classifier, Multilayer Perceptron (MLP), and LightGBM (LGBMC). LGBMC leads with the highest recall of 90.67%, showcasing its strong ability to identify positive cases. Gradient Boosting follows at 88.78%, while MLP and AdaBoost have slightly lower recall rates of 87.46% and 87.02%, respectively. LGBMC proves to be the optimal model for applications requiring high recall, while AdaBoost performs the least effectively. Additionally, receiver operating characteristic (ROC) curve analysis evaluates model performance and reliability. This technique enhances diagnostic accuracy and provides insights into feature importance, helping healthcare professionals understand the factors influencing predictions. Through XAI methodologies, the study bridges the gap between AI-driven diagnostics and practical clinical use, ensuring increased transparency, trust, and reliability in monkeypox detection.