Predicting Case Fatality of Dengue Epidemic: Statistical Machine Learning Towards a Virtual Doctor
Sudip Chattopadhyay, Amit K. Chattopadhyay, Elias C. Aifantis
Abstract
Dengue fever is a self-limiting communicable viral disease, transmitted through mosquito bites. Its Case Fatality Grade (CFG) varies across population due to variations in viral load, immunity of the patient, early diagnosis, and availability of high-end treatment facility. This study describes an initial effort to automate the process of Dengue CFG predictions. Two established Statistical Machine Learning (SML) algorithms, Multiple Linear Regressions (MLR) and Multinomial Logistic Regressions (MnLR), are combined to substitute the existing Deep Learning methods for clinical decision making. We consider a vector of eleven sign-symptoms (independent variables), each weighted between [0,1] on a 3-point scale - ‘Mild’ (CFG<=0.33), ‘Moderate’ (0.33<CFG< 0.66), and ‘Severe’ (CFG>0.66). Results show that both classifiers are effective in early screening with similar accuracy levels (68% for MLR versus 72% for MnLR) although precision levels are far superior with MnLR (88%) than MLR (61%). This study is a futuristic step towards Machine Learning (ML) aided clinical diagnostic paradigms, as an alternative to computationally intensive Artificial Intelligence.