Tracking dynamics change parameters of chaotic infectious disease outbreak with bifurcated time-series long short-term memory model

Abstract

In the dire need to develop a robust epidemiological surveillance model, this study aimed to determine if the dynamics change parameters of past infectious disease outbreaks would enable early accurate detection of future outbreaks. An optimized regular LSTM could not accurately learn the non-linear relationship of the epidemiological data in the peak stage of the pre-vaccination era measles outbreak, which hence influenced the decrease predictability at increasing time series. The result of this study shows that Pitchfork bifurcation of LSTM at 1.5 bifurcation point matches the dynamic change inherent in California's pre-vaccination era measles outbreak at the improved prediction performances of root mean squared error (RMSE) = 0.1684 for the 1962 outbreak, and root mean squared error (RMSE)= 0.2776 for 1964 outbreak. In conclusion, it was observed that the dynamics change parameters are dependent on the stochastic nature of the optimization algorithm engaged, and the chaotic nature of epidemiological data.

Keywords

• Time-series modeling
• Bifurcation theory
• Epidemiology
• Deep learning
• Bayesian optimization
• Long short term memory algorithm