Dual-grained directional representation for infectious disease case prediction

ABSTRACT

The uncertain infection transmission causes challenges in accurate disease prediction. Numerous methods have been proposed to capture the temporal pictures from past observations within equal time intervals, which are called single-grained time series. However, these methods are not suitable for capturing uncertain temporal dynamics from infectious disease time series, since the infectious diseases may propagate in the incubation period. To address this issue, this paper proposes a Dual-Grained Directional Representation (DGDR) to generate predictions, via consolidating the representations of an equal-grained time series and several fine-grained time series. Firstly, the proposed DGDR learns a transformed segmentation into three kinds of representations. And then those representations from both equal-grained data and fine-grained data are temporally consolidated to connect with outputs. Extensive experiments on two real infectious disease datasets are done to validate the proposed DGDR. Compared with the other twelve methods, MAE value is decreased by 31.5%, RMSE value is decreased by 29.9%, and R-2 value is improved by 87.6%. (c) 2022 Elsevier B.V. All rights reserved.