Prediction of the spread of Coronavirus Disease 2019 (COVID-19) is imperative for the efficient allocation and administration of national and worldwide healthcare. Traditional prediction models for COVID-19 pandemic are limited by their low accuracy because they assume homogeneous time-dependent transmission rates and isolate the study region in the absence of geographic features. To advance the prediction of COVID-19 spread, the parameters of the model should be refined via evolving understandings of the disease trajectory, transmission rates, and economic and social factors influencing infection. Hence, this study proposes a hybrid model combining the classic epidemic equations with a recurrent neural network (RNN) to predict the spread of the COVID-19 pandemic. The proposed model involves the number of patients who are Susceptible, Infectious, Recovered, and Deceased (SIRD) as time-dependent features and considers human mobility from neighboring regions as a spatial feature. This study derives a discrete-time function with the infection component of SIRD model for real-time application that avoids overfitting and is more efficient than other models. The proposed model was tested on a publicly available COVID dataset recorded from Italy. Experimental results show that it outperforms existing spatiotemporal models.