JMIR Publications

ABSTRACT

Background:

Changeful seasonal influenza activity in subtropical areas like Taiwan causes problems in epidemic preparedness. Taiwan Centers for Disease Control has built real-time national influenza surveillance systems since 2004. Except for timely monitoring, epidemic forecasting using the national influenza surveillance data can provide pivotal information for public health response.

Objective:

We aimed to develop predictive models using machine learning (ML) to provide real-time influenza-like illness (ILI) forecasts.

Methods:

Using surveillance data including ILI visits from emergent (from the Real-Time Outbreak and Disease Surveillance System), outpatient (from the National Health Insurance database) departments, and the records of patients with severe influenza with complications (from the National Notifiable Disease Surveillance System), we developed 4 ML models (autoregressive integrated moving average, random forest, supportive vector machine regression, and extreme gradient boosting) to produce weekly ILI predictions for the current week and next 3 weeks. We established a framework of the machine learning models and used an ensemble approach called stacking to integrate these predictions. We trained the models using the historical data from January 2008 and evaluated their predictive ability during 2015–2017 for each of the 4-week time horizons using Pearson’s correlation, mean absolute percentage error (MAPE) and hit rate of trend prediction. A dashboard web site was built to visualize the forecasts, and the results of real-world applications of this forecasting framework in 2018 were evaluated using the same metrics.

Results:

All models can accurately predict the timing and magnitudes of the seasonal peaks in 1-week (nowcast) (R = 0.802–0.965; MAPE: 5.2%–9.2%; hit rate: 0.577–0.756), 2-week (R = 0.803–0.918; MAPE: 8.3%–11.8%; hit rate: 0.643–0.747), 3-week (R = 0.783–0.867; MAPE: 10.0%–15.3%; hit rate: 0.669–0.734) and 4-week forecasts (R = 0.676–0.801; MAPE: 12.0%–18.9%; hit rate: 0.643–0.786), especially the ensemble model. In real-world implementation in 2018, the forecasting performance was still accurate in nowcasts (R = 0.875–0.969; MAPE: 5.3–8.0%; hit rate: 0.582–0.782) and remained satisfactory in 4-week forecasts (R = 0.721–0.908; MAPE: 7.6–13.5%; hit rate: 0.596–0.904).

Conclusions:

Using ML and ensemble approach can make accurate and real-time 4-week ILI forecasts and facilitate the decision making.