Sentiment analysis is vital for deciphering emotions in text, aiding in informed decision-making based on public sentiment. This study examines various machine learning models, focusing on deep learning architectures like GRU, LSTM, CNN-LSTM, and GRU-CNN. The analysis uses an Emotions dataset containing six categories—sadness, joy, fear, anger, love, and surprise—evaluating models based on accuracy, loss, and generalization. The GRU model stood out with a top accuracy of 91%. These results highlight the importance of selecting effective architectures to boost sentiment analysis performance and provide valuable insights for practitioners.