Forecasting influenza activity using self-adaptive AI model and multi-source data in Chongqing, China

Abstract

Background: Early detection of influenza activity followed by timely response is a critical component of preparedness for seasonal influenza epidemic and influenza pandemic. However, most relevant studies were conducted at the regional or national level with regular seasonal influenza trends. There are few feasible strategies to forecast influenza activity at the local level with irregular trends.

Methods: Multi-source electronic data, including historical percentage of influenza-like illness (ILI%), weather data, Baidu search index and Sina Weibo data of Chongqing, China, were collected and integrated into an innovative Self-adaptive AI Model (SAAIM), which was constructed by integrating Seasonal Autoregressive Integrated Moving Average model and XGBoost model using a self-adaptive weight adjustment mechanism. SAAIM was applied to ILI% forecast in Chongqing from 2017 to 2018, of which the performance was compared with three previously available models on forecasting.

Findings: ILI% showed an irregular seasonal trend from 2012 to 2018 in Chongqing. Compared with three reference models, SAAIM achieved the best performance on forecasting ILI% of Chongqing with the mean absolute percentage error (MAPE) of 11·9%, 7·5%, and 11·9% during the periods of the year 2014-2016, 2017, and 2018 respectively. Among the three categories of source data, historical influenza activity contributed the most to the forecast accuracy by decreasing the MAPE by 19·6%, 43·1%, and 11·1%, followed by weather information (MAPE reduced by 3·3%, 17·1%, and 2·2%), and Internet-related public sentiment data (MAPE reduced by 1·1%, 0·9%, and 1·3%).

Interpretation: Accurate influenza forecast in areas with irregular seasonal influenza trends can be made by SAAIM with multi-source electronic data.

Keywords: AI; Forecast; Influenza; Influenza-like illness; Multi-source electronic data.

Fig. 1

Time series of influenza-like illness percentages in Chongqing, China, 2012–2018.

Fig. 2

Estimation results of SAAIM in comparison of reference models. (A) The estimated ILI% values from SAAIM (thick red), comparing with the true CDC's ILI percentages (thick black) as well as the estimates from Lasso model with Baidu Index (blue), Lasso model with Baidu Index plus historical ILI% values(orange) and LSTM model (green) between the first week of 2014 and the last week of 2018. (B) The estimation error, defined as estimated value minus the CDC's ILI activity level. (C-E) Zoomed-in plots for estimation results in different study periods. (C) The 2014 flu season. (D) The 2015 flu season. (E) The real-time prediction of ILI percentages 1 week before official publication from March 25th, 2018 to December 30th, 2018. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 3

Importance analyses of different feature groups. SAAIM was constructed with four kinds of features: historical ILI, weather, sentiment and time. (A) The estimates of SAAIM without the climate features (blue), the public sentiment features containing Baidu Index and Weibo (green), without the historical ILI features (orange) are drawn. The estimated ILI% values of SAAIM with all features (red) and the true CDC's ILI activity level (black) are shown as references. (B) The estimation error, defined as estimated value minus the CDC's ILI activity level. (C-E) Zoomed-in plots for estimation results in different study periods. (C) The 2014 flu season. (D) The 2015 flu season. (E) The real-time prediction of ILI percentages 1 week before official publication from March 25th, 2018 to December 30th, 2018. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)