Automated machine learning models to assist in COVID-19 surveillance, diagnosis, and intervention

Date

2021-12

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

Since its first outbreak, Coronavirus Disease 2019 (COVID-19) has been rapidly spreading worldwide and caused a catastrophic global pandemic. How to prevent COVID-19 spread efficiently and effectively has been an urgently important challenge for the governments and policymakers all around the world. With the largely available public resources, how to improve our knowledge and gain new insights about COVID-19 is critical for effective interventions and adaptations to mitigate the associated risks. During the last decades, machine learning (ML) based methods have shown unprecedented success in various domains, such as reliable analysis of prediction, classification, and natural language processing. In this dissertation, we proposed and developed a series of reliable and automated machine learning algorithms to assist COVID-19 surveillance, diagnosis and intervention.

Specifically, we first investigate the relation between Google search trends and the spreading of COVID-19 over countries worldwide, to predict the number of cases. Our results show that Google search trends are highly associated with the number of reported confirmed cases, where the Deep Learning approach outperforms other forecasting techniques. Secondly, we propose a novel SIHR model, where we introduced a new hospitalization state to the classic SIR pandemics mode, to model the spread of this highly infectious diseases for understanding epidemic duration, the number of infected people throughout the duration, and the peak number of infected people etc. Thirdly, we developed a question answering system to answer medical questions regarding COVID-19 related symptoms using Wikipedia articles and other source cloud dataset. Besides, we built a knowledge graph framework which is able to extract information accurately and efficiently from a myriad of clinical test results articles, for instance, to find the supporting and/or contradictory references of certain drugs. Lastly, a visualization platform is developed to provide decision-makers with accurate data-driven representations in an easy to understand format that enables them to make more timely and cost-effective preparation and response plans. In summary, we developed a series of automated machine learning solutions to assist the COVID-19 related researches, ranging from spread modeling, dynamic analysis, containing and disease intervention, based on data cloud sourcing methodology.

Description

Keywords

Machine Learning, COVID-19, Prediction, Deep Learning, Knowledge Graph, Epidemiological Modeling

Citation