Regression Chain Model for Predicting Epidemic Variables.

Real-time detection and forecasting of disease dynamics is critical for healthcare authorities during epidemics. In this paper, we report a systematic investigation into the possibility of predicting three epidemic variables, viz., peak day, peak infections, and span of the epidemic using the Regression Chain Model. We construct a dataset, EpiNet, using 35K synthetic networks of varied sizes and belonging to three network families. The dataset consists of five network features and three target variables obtained by simulating the SEIR epidemic model on the networks. We train Regression Chain Model (RCM) using four popular machine learning algorithms to predict the target variables. The model generally performs fairly well for peak day and peak infections, but the performance degrades for the span variable. Our preliminary investigation motivates further inquiry into the use of RCMs to replace computationally expensive epidemic simulations on larger networks.