A Framework For Multi-Class Prediction Of Dengue Using Early Clinical And Laboratory Indicators

Dengue is a mosquito-borne disease. Effecting nearly 400 million people world-wide with its three severity  levels and four serotypes it has devastated more than 100 countries. Since dengue’s first case in 1994, Pakistan saw its four surges during year 2005, 2011, 2014 and 2019. Having no vaccine or cure the only safeguard against this virus is early detection of its disease progression. There have been many studies done in the past but mainstay remained the hospitalization of patients for monitoring and prediction. This puts additional burden on the healthcare departments during bulk patient management. Use of apt clinical and laboratory indicators are vital for understanding possible danger, the patient may have for appropriate treatment. This study will use machine learning algorithms for evaluating best clinical and laboratory indicators suitable for classifying dengue severity levels and its serotypes. This research also proposes the use of deep learning and ensemble methods to classify dengue severity level and its serotypes for better management of patients and at the same time emergency healthcare services are not disturbed. The models have been evaluated using 10-fold cross validation. The results show that ensemble machine learning algorithm Random Forest perform better against neural network with accuracy of 89\% while we achieve 85\% accuracy using MLP Classifier.