PNNL

Abstract

Despite the complex and unpredictable nature of pathogen occurrence, substantial efforts have been made to better predict infectious diseases (IDs). Following PRISMA guidelines, we conducted a systematic review to investigate the advances in ID prediction capabilities for human and animal diseases, focusing on Machine Learning (ML) and Deep Learning (DL) techniques. Between January 2001 and May 2021, the number of relevant articles published steadily increased with a significantly influx after January 2019. Among the 237 articles included, a variety of IDs and locations were modeled, with the most common being COVID-19 (37.6%) followed by Influenza/influenza-like illnesses (8.9%) and Eastern Asia (32.5%) followed by North America (17.7%), respectively. Tree-based ML models (38.4%) and feed-forward DL neural networks (26.6%) were the most frequent approaches taking advantage of a wide variety of input features. Most articles contained models predicting temporal incidence (66.7%) followed by disease risk (38.0%) and spatial movement (31.2%). Less than 10% of studies?addressed the concepts of uncertainty quantification, computational efficiency, and missing data, which?are essential to operational use and deployment. Our study summarizes the broad aspects and current status of ID prediction capabilities and provides guidelines for future works to better support biopreparedness and response.