Machine Learning-based Classification of COVID-19 Preventive Behaviors Among University Staff and Students in Chiang Mai, Thailand

Abstract

The COVID-19 pandemic has exposed public health management to new challenges. Most importantly, there is a need to assess risk and deploy limited resources quickly in response to emerging disease outbreaks. Previous research has shown that self-reported survey data can accurately predict infection risks. However, even this can be slow and cumbersome to implement. Herein, we explored a new approach. We collected risk assessment survey data from 1,266 members of the Chiang Mai University community. We found that students were generally at a higher risk of developing COVID-19 than faculty members. However, we were interested in knowing whether this difference was due to heterogeneous risk profiles or whether students and faculty exhibited systematic risk profiles that were largely homogeneous within each group. To assess this, we trained machine learning models to classify participants as students or faculty group members. This model achieved an accuracy of 95% in a test data set, confirming its ability to categorize group membership and suggesting that risk profiles must have been relatively homogenous within each group. This result suggests that public health decision makers can confidently make decisions to deploy resources to help large societal groups based on survey data collected from relatively small samples.