Early Detection and Prediction of Zoonotic Disease Events Using Event-Based Surveillance and Machine Learning

The increasing frequency and scale of zoonotic disease outbreaks in recent years have amplified the need for better disease surveillance initiatives. However, building and maintaining such surveillance systems is a challenge, especially in low or middle-income countries. In this study, we examined the potential of open-source data and Event-Based Surveillance (EBS) in the early detection and prediction of zoonotic diseases. We extracted key disease event-specific information from the open-source data using a newly developed disease event taxonomy. We conducted a PRISMA systematic review to study the advances in using Machine Learning (ML) in infectious disease prediction over the past two decades. Finally, we utilized EBS along with ML and transfer learning (TL) techniques to predict an emerging zoonotic disease i.e., Kyasanur Forest Disease (KFD) cases in humans under resource and data-limited settings. Using news media as a form of EBS, we identified zoonotic disease events well in advance of the official reporting in Kenya. Rift Valley fever and anthrax were the most frequently reported zoonotic diseases. The systematic review showed an increasing trend in the use of ML techniques for infectious disease prediction in the past two decades with tree-based ML and feed-forward neural networks being the most frequently used techniques. Temporal models predicting highly contagious and zoonotic diseases in humans were particularly popular. However, prediction models in resource-scare settings across regions and diseases were underrepresented. For KFD, using EBS along with weather data increased the predictive performance of time-series ML models when compared to using weather data alone. The TL enabled accurate and timely prediction of KFD cases in new outbreak regions with limited epidemiological data. Our study demonstrates the potential of novel data sources such as EBS and advanced ML approaches in increasing disease prediction capabilities in resource-scarce situations for better-informed decisions in the face of emerging zoonotic threats.