COVID-19 (Coronavirus) is a pandemic situation and it affects the social and economics of the country. Hence, there is a need for robust prediction model that can ensure better prediction outcomes. Early diagnosis and finding the severity of diseases are the major concerns of the medical experts. An automated model is essential for predicting the severity of the COVID-19 and the diagnosis will assists the experts. Deep learning (DL) models are efficient mechanism that can able to process the massive data which is in different formats like clinical symptoms. This work presents the DL model to analyze the medical data consists of 65,000 patients records and 26 features. Here, the optimal features are selected by the optimizer called Enhanced sky driver optimization (ESDO). With the selected features, the DL model Long short Term memory (LSTM) based AutoEncoder (AE) is used to screen the severity of COVID-19 patients. The performance of the proposed LSTM based AE is compared with other DL models and achieved better accuracy of. Further, this model predicted the severity cases as mild, moderate and severe. The medical experts can use these outcomes for finalizing the type of medical treatment that has to be given to the affected patients on time.