Forecasting of Influenza-like Illness Incidence in Amur Region with Neural Networks

Abstract

Influenza-like illness (ILI) incidence forecasting strengthens disease control and prevention. The virus, host and host behavior factors influencing ILI outbreaks have been studied for decades. A range of statistical and machine learning forecasting methods was developed. These novel machine learning methods require inclusion of a proper factor set based on a systematic research. The conventional forecast evaluation metrics such as Mean Absolute Error (MAE) do not adequately reflect the epidemiological requirements and shall be replaced with tailored evaluation criteria. This paper discusses selection of the main influencing factors based on the recent epidemiological research, and proposes new epidemiological forecast evaluation criteria to asses early-warning power of the short-term forecasting model. It describes development of a prediction model based on a Long-Short Term Memory (LSTM) neural network. The model was implemented, trained, validated and tested on the 2007–2018 historical data set and compared to Local Autoregressive Models, Autoregressive Integrated Moving Average, and Multivariate Regression methods.