One of the biggest challenges in higher educational institutions is to avoid students’ failures. Globally student dropout is a serious issue. Risk of dropouts can be identified at an earlier stage using machine learning classifiers, as they have gained more popularity in both academia and industry. The research team suggests that early prediction facilitates educators and higher education administrators to take necessary measures to prevent dropouts. Data for the research were collected from 530 Indian students when they were engaged in online learning during pandemic crisis. This research work involves two phases. In first phase, hybrid ensemble strategy is focused that integrates two powerful machine learning algorithms namely Random Forest (RF) and eXtreme Gradient Boosting (XGBoost) for early at-risk prediction. The result is a fast procedure for classification of at-risk students which is competitive in accuracy and highly robust. Prediction models are developed using ensemble learning, furthermore ensemble models are combined into a single meta-model, which provides best outcomes to enable higher education institutions for predictive analysis. Moreover, it correctly classified students’ at-risk regarding accuracy, precision, recall and F1-score with values of 93%, 91.52%, 96.42% and 93.91% respectively. In second phase, prediction model is deployed by creating a web application using. Net framework to sense students’ sentiments using Azure cognitive services text analytics (Application Programming Interface) API for detecting cognitive behavioral outcomes in online learning environment.