Machine learning models to predict the risk of malaria outbreaks using balanced climatic and non-climatic data

Objective: To predict malaria outbreaks early and accurately using machine learning models when data presents challenges such as imbalance. Methods: To predict the risk of malaria outbreaks, we applied six machine learning models—including naive Bayes, logistic regression, random forest, K-nearest neighbors, decision tree, and extreme gradient boosting—to a dataset of 8000 samples with 17 climatic and non-climatic features. To address class imbalance, we used the Synthetic Minority Over-sampling Technique algorithm. The models were evaluated using 10-fold cross-validation repeated 10 times to ensure robust validation. Results: Extreme gradient boosting achieved the highest F1-score, improving from 0.926 on imbalanced data to 0.991 after balancing. While random forest, naive Bayes, and decision tree also improved their performance when balancing data with Synthetic Minority Over-sampling Technique (increasing F1-score from 0.893 to 0.931, from 0.765 to 0.962, and from 0.785 to 0.812, respectively), logistic regression and K-nearest neighbors did not show any improvement. Conclusions: These findings demonstrate that integrating balanced climatic and non-climatic data with advanced machine learning can improve malaria outbreak risk prediction and support public health strategies.