World Health Organization (WHO) publishes infectious disease outbreak news with information such as locations, dates, and cases. The information is often under-utilized, as manual reading is required to act upon them.
What it does
Outbrake is an artificial intelligence system that can read disease outbreak news and extract the key information.
How I built it
Words that correspond to key entities such as disease, location, date and case are labeled like this.
The training data is fed to an encoder/decoder architecture with attention mechanism to produce a machine learning model.
Challenges I ran into
Detecting number of cases is difficult due to many variations such as no new cases, one confirmed case, or 1000 probable cases.
Dates are also difficult to detect due to variations such as 10 April 2020, or 15 to 20 March 2020.
Accomplishments that I'm proud of
Open source. Fighting epidemics and pandemics require effort from everyone.
Fair and inclusive. Data is trusted and highlights third-world and poor countries often neglected by mainstream media.
User privacy is protected by data aggregation and anonymization.
Disease, locations, dates, and number of cases are extracted in real time from WHO disease outbreak feed.
What I learned
COVID-19 gets all the attention today. But infectious diseases like ebola and MERS-CoV are ongoing threats, especially in third-world countries. Locations where these other diseases are active should take extra precautions. Extra resources like personal protective equipment should be set aside in these locations.
What's next for Outbrake
Outbrake lays the foundation for future-state AI systems that prevent pandemics. An example is a case tracking application that uses data extracted from disease outbreak news feeds. Users can see case trends by location. Another use case is an early-warning system for more severe infectious diseases such as MERS-CoV.