Predicting the length of stay (LoS) plays a crucial role in resource management both from the patient as well as hospital’s perspective. From the hospital’s perspective, it empowers them to improve patient care, reduce costs for the patient’s family, and allocate hospital resources while on the patient’s side, it empowers their families to plan the costs and the period in which the patient has to stay at the hospital. LoS prediction is an important parameter in several diseases like COVID-19, community acquired pneumonia (CAP), type-2 diabetes mellitus and hypertension, as these lead to increased hospital admissions and burden on hospitals. In this study, we have developed an automated framework named to predict the LoS in hospital for COVID-19 infected patients. A COVID-19 dataset was particularly chosen because this particular epidemic had created an unprecedented strain on our medical facilities since its outbreak in 2019. We have used a Kaggle challenge dataset, namely AV: Healthcare Analytics II dataset for our study. The main task of this challenge was to accurately predict LoS for COVID-19 admitted patients and categorize them into 11 different classes. We have developed an ensemble model, called COVLoS-Net comprising a customized convolutional neural network (CNN), CatBoost, Random Forest, Support vector machine (SVM), k-nearest neighbors (KNN) as the base models to predict the LoS for COVID-19 infected patients requiring hospital admission. The predictions generated by our model on the test set of the challenge yields an accuracy of 43.94, and this score achieves the Rank 1 as evaluated by the challenge organizers. Further, we have also compared our model with the state-of-the-art models to show that our results surpassed the existing models for LoS prediction.