Public announcements and mask use during COVID-19 bridging regression and machine learning.

The effectiveness of public health policies has a major impact on society. Throughout the COVID-19 pandemic, policy-makers worldwide introduced various public health measures, such as social distancing and mask-wearing, to address the severe threat posed by the disease. Despite the importance of these issues, there has been no direct, visual evidence that mask-related policies led to the appropriate mask-wearing behaviour of the general public. Here, we show that regulations and public transport announcements encouraged correct mask-wearing behaviour during the COVID-19 pandemic. Importantly, changes in announcement and regulation contents led to heterogeneous effects on behaviour. Our study contributes the first visual open-source empirical analysis of reactive human behaviour during the COVID-19 pandemic, investigating how compliant a general population is to mask-wearing policy. We combine object-detection-based convolutional neural networks, regression analysis and multilayer perceptrons to analyse visual data of the Viennese public during 2020. By comparing the predictive power of regression analysis and neural networks, we demonstrate that the neural networks provide more accurate predictions of population response during the COVID-19 pandemic. Furthermore, our use of regression modelling allows us to uncover possible causal pathways underlying societal behaviour. Since our findings underscore the importance of appropriate communication content, our work will facilitate more effective non-pharmaceutical interventions in future public health crises. Lastly, we demonstrate that the use of regression modelling and neural networks need not be mutually exclusive in the study of complex systems but can instead complement each other.