Finnish researchers used AI and machine learning algorithms to identify the number of flu-related hashtags and images on Instagram to predict a flu epidemic
As the debate over the impact of social media on mental health continues, researchers in Finland have discovered a physical health benefit to networking by using Instagram to predict when a flu epidemic will happen.
By comparing the number of Instagram posts that related to flu-like symptoms and the Finnish national statistics on public health, Oguzhan Gencoglu and Miikka Ermes were able to successfully predict flu outbreaks.
Mr Gencoglu, who along with his colleague is a researcher in the medical faculty at the University of Tampere in Finland, said: “This is the first paper to apply visual analysis to predict epidemics, as far as we know.
“We selected seven keywords and phrases, such as ‘flu’ or ‘sore throat’. We researched all those hashtags and crawled the data on all those Instagram posts to look at the text and the image.
“We did weekly counts of those hashtags and analysed the visual content using a recently developed deep learning AI, which was able to count the weekly images that featured medicine boxes, pills or mugs of ginger tea – which we realised people shared a lot.”
The predictive model was trained from 317 weeks of Instagram data and was then used to forecast the number of of official influenza-like illness incidents across the country.
When compared to the official weekly flu counts on the Finnish national healthcare register, the AI model had a mean error of 11.3 incidents per week and had a correlation coefficient of 0.963 – where one is a perfect linear relationship.
When making predictions two weeks into the future, the model reached a correlation of 0.903, showing that Instagram can be a valuable source of information for healthcare professionals when predicting influenza epidemics.
Why is it important to predict a flu epidemic?
By the end of December last year, the number of influenza patients hit epidemic levels in Finland, according to the country’s National Institute for Health and Welfare, with 2,563 diagnoses.
Mr Gencoglu believes that having a model to predict flu epidemics from social media posts is of “significant importance”.
He sees a number of applications for his discovery, such as optimising the number of healthcare professionals working when an epidemic is likely to occur and using the early notification to create interventions around cities to stop the infectious disease spreading.
The prediction tool could also be applied to other types of contagious diseases by training the deep learning algorithms on different data.
Mr Gencoglu said: “We wrote this paper to open up the field and show the opportunity of using Instagram as a valuable data source for spotting epidemics.
“It also shows a new method for analysing images through new AI algorithms.”
Other social media channels could be used to monitor national health.
However, Instagram held a couple of additional benefits over other social media channels because of the likelihood of people to share such images and the fact the majority of accounts are public.
Mr Gencoglu added: “Instagram is extremely popular nowadays so this gives it an advantage in terms of the amount of data available.
“These algorithms are quite data hungry so you need a lot of data to train them.”