Improving malaria prediction with ensemble learning and robust scaler: An integrated approach for enhanced accuracy

Abstract

Mosquito bites are the primary transmission method for malaria, a prevalent and significant health concern worldwide. In the context of malaria incidence, Indonesia is the second most affected country after India. According to the Ministry of Health's report, Papua Province reported 216,380 malaria cases in 2019. Additionally, East Nusa Tenggara and West Papua said 12,909 and 7,029 points, respectively, reflecting the substantial national burden of this disease. Predicting malaria occurrence based on symptomatic presentation is a crucial preventive strategy. Machine learning models offer a promising approach to malaria prediction. This study focused on malaria detection by using patient data from Nigeria. This research proposes a detection system utilizing the Ensemble method, such as Decision Tree, Random Forest, and Bagging. This study also employing Robust Scaler for effective normalization and integrating K-fold cross-validation to enhance model robustness. Various experiments were conducted by systematically varying K values and the number of decision trees to ascertain the most effective hyperparameters yielding the highest accuracy. The findings indicate that the optimal accuracy 82% is achieved at a K value of 20, showing comparable accuracies across different decision tree quantities, underlining the robustness of the employed method. This research significantly advances malaria detection strategies, offering valuable insights into the effective deployment of machine learning in healthcare decision-making.