Machine Learning based Malaria Prediction using Clinical Findings

Abstract

Even today, Malaria is the most deadly disease in Asia and sub-Saharan Africa and particularly in Senegal. This is mainly due to inadequate medical care support with frequent late and error-diagnoses by medical professionals. Besides, mostly used diagnostic standards such as the rapid diagnostic test is not fully reliable. With the development and widespread acceptance of automated systems in the healthcare system, machine learning algorithms can support medical professionals in their decision-making procedure. An experimental analysis of different machine learning techniques to predict Malaria is proposed in this work. These techniques attempt to determine whether or not a patient suffersfrom Malaria using various clinical findings like signs and symptoms. The algorithms' efficiency has been thoroughly validated and analysed over two actual data sets of malaria patients' taken from Senegal. The results obtained show that Random Forest, Support Vector Machine with Gaussian Kernel and Artificial Neural Networks are promising and offer the best overall accuracy to predict the appearance or not of the disease with precision, recall and F1-score at least equal to 92%, 85% and 89% respectively on both datasets on which they outperform the Rapid Diagnostic Test.