A multi-prefecture study applying multivariate approaches for predicting and demystifying weather data variations affect COVID-19 spread

Since the first cases were reported in Wuhan, China in December 2019, the massive SARS virus, known as COVID-19, is spreading at an alarming rate and is endangering the world. The whole world is now affected by this terrible epidemic including Pakistan. Because it provides a rich resource for studying the variants affecting the COVID-19 contagion, the spread of COVID-19 in Pakistan makes it necessary to assess the link between COVID-19 cases and weather parameters for management and policymakers. The proposed study uses a multivariate approach to demystify the impact of meteorological conditions on COVID-19 contagion, where in a case study, different regions of Pakistan are chosen. The multivariate techniques used in the proposed approach comprise linear multiple regression, linear stepwise regression, multiple adaptive regression splines, and loess regression. We also implement spline regression models for deeper analysis. A machine learning regressor was used to validate the results of the spline curve regression model. The performance of machine learning models and splines is measured using different performance metrics. Experiments show that weather parameters such as temperature and humidity are prominent features in predicting the spread of COVID-19. The spline curve results show that, except for Baluchistan, during the first wave, the temperature was positively correlated with daily confirmed cases, and during the second wave, the temperature was negatively correlated with daily confirmed cases in all other provinces. Another finding of the study is that when data from March 2020 to February 2021 were included, the humidity was inversely associated with mortality and confirmed cases in all provinces. Thus, increased humidity inhibits the spread of COVID-19 and reduces mortality.