AI-driven diagnosis of Lassa fever: Evidence from Nigerian clinical records

Abstract

BACKGROUND: Neglected Tropical Diseases (NTDs), particularly Lassa fever, remain a significant public health challenge in Nigeria, often presenting with symptoms similar to malaria. These similarities contribute to misdiagnoses, delayed treatments, and increased mortality. The need for rapid and accurate disease differentiation has created an opportunity for machine learning applications in medical diagnostics. METHOD: This study developed an ensemble machine learning model to detect Lassa fever and distinguish it from malaria using clinical datasets collected from the Infectious Disease Hospital, Akure, and the Benue State University Teaching Hospital, Makurdi. The dataset, comprising confirmed Lassa fever and malaria cases, underwent preprocessing steps including data cleaning, handling missing values, balancing via SMOTE, and feature selection using ANOVA. Three base classifiers: Support Vector Machine (SVM), K-Nearest Neighbours (KNN), and Multi-Layer Perceptron (MLP), were combined using a hard voting ensemble technique. Model performance was evaluated using accuracy, precision, recall, and F1-score. RESULTS: The ensemble model outperformed the individual classifiers, achieving an accuracy of 98.7 %, precision of 98.3 %, recall of 100 %, an F1-score of 99.1 %, and ROC-AUC of 96.88 %. These results represent a significant improvement over existing approaches, with the ensemble model demonstrating 8.7 % higher accuracy compared to the best individual classifier (KNN at 90 %) and substantially outperforming traditional diagnostic methods that typically achieve 60-70 % accuracy in differentiating Lassa fever from malaria in resource-limited settings. These results indicate a robust capacity for differentiating Lassa fever from malaria based on symptomatology. CONCLUSION: The ensemble learning approach demonstrated high effectiveness in improving disease detection accuracy, making it a practical tool for early diagnosis and clinical decision support in resource-limited healthcare settings. Its deployment could significantly reduce misdiagnosis and enhance NTD surveillance in Nigeria.