Epicasting: an ensemble wavelet neural network for forecasting epidemics (Q6057959)

This paper presents an ensemble wavelet neural network for forecasting epidemics, which does not have growing variance over time and exhibits good long-range forecast ability for epidemic datasets. The proposed formulation can handle nonlinear, non-stationary and long-range dependency of real-life data. Some theoretical properties, including asymptotic stationary, ergodicity, irreducibility and learning stability, have been examined. The paper explores the global characteristics of fifteen real-world infectious disease datasets including influenza, malaria, dengue, and hepatitis B from a range of areas. It is shown that the proposed framework can generate a better long term forecast and it outperforms many forecasters on average. A non-parametric test is used in the robustness study of the forecast.