LightGBM and Voting Classifier: Top Performers in Supervised Classification for Vector-Borne Diseases in Hauts-Bassins, Burkina Faso

Abstract

Vector-borne diseases remain a major public health challenge, particularly in low-income countries where access to laboratory diagnostics is limited. This study evaluates the performance of 11 supervised learning models for classifying vector-borne diseases, using a dataset of 300 patient records from Burkina Faso, where malaria accounts for approximately 79% of cases. To address class imbalance, the Synthetic Minority Oversampling Technique (SMOTE) was applied. The results indicate that LightGBM and Voting Algorithm stand out as the best-performing models among those tested. Specifically, LightGBM achieved the highest accuracy on balanced datasets, with an accuracy of 98.3% and an F1-score of 98.2%. Meanwhile, the Voting Algorithm performed best on imbalanced datasets, achieving an accuracy of 86.44% and an F1-score of 83.40%. These findings highlight the importance of selecting an appropriate model based on dataset characteristics. This study emphasizes that the accuracy of vector-borne disease prediction can be significantly improved by exploring additional machine learning models. Regardless of whether the dataset is balanced or imbalanced, tailored approaches can optimize classification performance. Finally, these findings offer new perspectives on the application of artificial intelligence to enhance disease diagnosis in resource-limited settings.