Weekly ILI patient ratio change prediction using news articles with support vector machine

Area of Science:

• Public Health
• Epidemiology
• Computational Biology

Background:

• Influenza poses a significant global health threat, necessitating accurate detection of infection patterns.
• Predicting influenza spread is challenging due to its sporadic and rapid nature, and limited data availability.
• Identifying novel data sources and developing robust prediction models are crucial for forecasting influenza incidence.

Purpose of the Study:

• To evaluate the utility of internet news articles as a data source for predicting influenza spread.
• To assess the accuracy of a support vector machine (SVM) model in forecasting weekly influenza-like illness (ILI) patient ratios using news text data.

Main Methods:

• Collected 7,769 infectious disease-related internet articles from Hong Kong's Centre for Health Protection (2004-2018).
• Utilized news text data from 2013-2018 to predict weekly ILI patient ratio changes.
• Employed a support vector machine (SVM) model to analyze patterns in news articles and predict influenza variance.

Main Results:

• The SVM model achieved a mean accuracy of 86.7% in predicting increases or decreases in the weekly ILI patient ratio.
• The model demonstrated a root mean square error of 0.611 for estimating the weekly ILI patient ratio.
• News text data proved effective in capturing patterns related to influenza spread.

Conclusions:

• News articles offer a viable alternative data source for influenza prediction, overcoming limitations of conventional data.
• This approach can help estimate both the direction (increase/decrease) and magnitude of ILI patient ratios.
• Further research into leveraging news articles for influenza forecasting is warranted due to promising performance.