In the last few months, the corona virus crises have changed the world completely. In India, looking at the horrific condition and rate of increase of COVID-19 positive cases, the central government has imposed strict lockdown from dated March 24 throughout the country. At the same time, with citizens quarantined and grinding halt in industrial activities has led to some inadvertent benefits and change in air pollution level is one of them. The present study is undertaken with an aim to predict and assess the change in PM2.5 level of the Kanpur city using MODIS Aerosol Optical Depth (AOD) data and meteorological parameters during various stages of COVID-19 lockdown. Subsequently, four different methods such as linear regression (LR), multilinear regression (MLR), artificial neural network (ANN), and a hybrid method were used and compared for the prediction of PM2.5. The best prediction method was further used for the spatial mapping of PM2.5 at different stages of lockdown (pre-lockdown, during lockdown and post lockdown). The result of cross-validation shows that the hybrid method i.e. (MLR+ANN) outperforms all other methods and gives the highest coefficient of determination R2 value of 0.961 followed by ANN, MLR, and LR with 0.895, 0.246, 0.016. The spatial mapping of PM2.5 during the lockdown shows a significant reduction of 26% in PM2.5 concentration in comparison to pre lockdown. The methodology developed from the current study can also be used in other regions for predicting and analyzing the spatio-temporal variation of PM2.5.