With the growing popularity of social networking sites in recent years, people now have a forum to share business proposals, recommendations, and advertisements. This research work focuses on emotions expressed by humans in social media texts to comprehend and analyze emotions such as joy, anger, sadness, and fear. The experiments were carried out with the fundamental deep learning models such as CNN (Convolutional Neural Networks), LSTM (Long Short Term Memory), Bidirectional LSTM (BiLSTM), Gated Recurrent Units (GRU), and Bidirectional GRU (BiGRU) for different levels of classifications of emotions in the textual data. The novelty of the proposed approach is the application of the ensemble model with a transformer learning model for the classification of emotional texts. The proposed approach is significant across the various fields that rely on interpreting the emotional cues found in social media texts. The results obtained with the proposed approach are better than other conventional methods. Our proposed work achieves an accuracy of 96.02% in the identification of different levels of emotions in social media texts such as Twitter and also provides better results on different standard datasets such as SemEval 2018, Sem Eval 2019, and MELD.