Machine learning based prediction of malaria outbreak using environment data in Rwanda

Abstract

Malaria is still one of the common diseases that cause a threat to human population globally, 2019 report of the World Health Organization (WHO), indicated that an estimated 228 million malaria cases were found in 2018 and in 2017 the estimated malaria cases were 231 million. Many studies have highlighted the environment factors as the contributing factors to the variability of malaria prevalence in different regions. Different study revealed that climate factors including temperature, humidity and rainfall should play an important/leading role in the model prediction of malaria outbreak. Furthermore, there is a need for intelligent predictive systems using machine learning techniques which can predict the malaria outbreak, based on historical data on malaria and environment factors, this study was based on building a predictive model of malaria outbreak which should be used as a warning system to health care providers, hospitals, and other health institutions to give a warning on an occurrence of malaria outbreak based on meteorological data. Historical data from 2016 to 2019 of malaria cases from Rwanda Biomedical Center (RBC) and meteorological data including maximum and minimum temperature, rainfall, elevation, latitude ,and longitude from Rwanda Meteorological Agency of ten districts including Nyarugenge, Kicukiro, Nyamagabe, Huye, Gicumbi, Musanze, Karongi, Rusizi, Kayonza, and Nyagatare that constitute the total of 2080 observations have been used to fit six machine learning algorithms including Decision tree, Random Forest, Naïve Bayes, Support vector machine, K-nearest neighbour, and Logistic regression thus the best algorithm model will be selected based on performance metrics of Accuracy, Precision, Recall, F-score and ROC have been applied on each machine learning algorithms to evaluate their performance. Among those classifiers, Random Forest comes with high performance compared to others with more than 90% in all evaluation metrics; it has shown the accuracy of 90.75%, F-score of 90.73%, Precision of 90.69% and Recall at 90.88%. However, the classifiers also have shown the high performance except for Support vector machine which shown only around 60% in all evaluation metrics, but other classifiers scored above 70% on the evaluation metrics. According to this study for malaria outbreak, it is recommended to use Random Forest for malaria prediction. The use of machine learning algorithms models for prediction of malaria outbreak proved to be used as alarming system for health-based, health care providers, and health practitioners thereby they can be well prepared and set up the new prevention measures.