Study shows how social media posts and AI can detect heat stroke risks in real time

Japanese researchers report

Researchers in Japan have demonstrated the potential of combining social media posts with deep learning models to detect heat stroke risks in real time, offering a new tool for public health surveillance. This innovative approach could help mitigate the impact of heat stroke during extreme heat events, which are becoming more frequent due to climate change.

Heat stroke remains a significant health threat, particularly in the face of rising global temperatures and increasing heatwaves. Vulnerable populations, including older people, children, and those with pre-existing conditions, are at heightened risk. In response to this challenge, researchers have been exploring new ways to detect heat stroke events early, enabling timely intervention.

While social media platforms like X (formerly known as Twitter) have previously shown promise in detecting infectious diseases in real-time, their use in tracking heat stroke risks has not been explored until now.

A team of Japanese researchers, led by Professor Sumiko Anno at Sophia University’s Graduate School of Global Environmental Studies, has pioneered using social media data combined with advanced machine learning models to detect signs of heat stroke in urban environments.

The study, published in Scientific Reports on January 4, 2025, focuses on Nagoya City, Japan, and harnesses transformer-based models, such as BERT, RoBERTa, and LUKE Japanese base lite, in combination with a machine learning model called Support Vector Machine (SVM).

The team analysed approximately 27,040 tweets containing the word “hot” in Japanese, collected over five years using the X API. The researchers then preprocessed the data and applied deep learning techniques to train models capable of identifying tweets related to heat stroke incidents.

The deep learning models were evaluated using key performance metrics such as accuracy, precision, recall, and F1-score. Among the models tested, LUKE Japanese base lite outperformed the others, achieving an accuracy rate of 85.52%. BERT-base followed closely with 84.04%, while RoBERTa-base reached 83.88%. The SVM model, used as a baseline, achieved the lowest accuracy of 72.73%.

The researchers also demonstrated the potential for real-time event-based surveillance using time-space visualisations and animated video. The study showed how social media data could serve as an early warning system for heat stroke risks by mapping emergency medical evacuations and linking them to geo-tagged tweets. This innovation allows faster identification of heat-related health threats and can help authorities deploy resources more efficiently in high-risk areas.

Professor Anno noted: “By leveraging social media posts, we can enhance public health surveillance systems and facilitate the early detection of heat stroke risks. Our findings underscore the importance of real-time data monitoring to address the health challenges climate change poses.”

This research highlights the potential of combining social media data with transformer-based pre-trained language models to monitor public health risks. LUKE’s superior performance in detecting heat stroke-related tweets suggests it could be a valuable tool for public health surveillance during heatwaves.

Furthermore, integrating social media data with emergency response systems shows how technology can improve the speed and accuracy of early detection tools for extreme weather events.

The implications of this study are far-reaching, especially as climate change continues to intensify the frequency and severity of heat waves. The team’s findings could help shape the development of future early warning systems that protect public health.

The researchers plan to expand this approach by establishing an early warning system for heat stroke in Aichi Prefecture, eventually creating a nationwide alert system across Japan. This system would rely on the collaboration of local authorities to gather data on heat stroke cases and conduct spatiotemporal analyses across various prefectures.

Professor Anno believes this methodology could be adapted to monitor other public health threats, such as emerging infectious diseases. “Our approach can be extended to track new and reemerging diseases, broadening its application in public health surveillance,” she concluded.

Integrating social media and advanced machine learning models represents a significant step forward in early detection and response to heat stroke risks. As the world faces the growing challenges of climate change, this research provides a promising solution to safeguarding public health during extreme heat events.

Established as a private Jesuit-affiliated university in 1913, Sophia University is one of the most prestigious universities in Tokyo.

Image: The Sophia University study showed how social media data could be an early warning system for heat stroke risks. Credit: Ketut Subiyanto