E-Commerce reviews play an important role in deciding whether to buy or not a product or to authenticate the product from a customer’s perspective. Currently there are 237 million native Bengali speakers, and the number of e-Commerce sites is also growing day by day. Sentiment analysis is very important for understanding consumer reviews and improving online buying experiences. Although there has been much work done on sentiment classification in English and other languages but not much in Bangla because of its complex structure and the scarcity of labeled datasets. So, we are working on the classification of sentiment in Bangla e-commerce reviews, particularly from Daraz and Pickaboo, 2 most popular e-commerce platforms of Bangladesh. 10,000 Bangla-only reviews were preprocessed by eliminating emojis, punctuation, and digits from a dataset of 78,130 reviews. We implemented three deep learning models which includes BanglaBERT, Long Short-Term Memory (LSTM) based model, and Gated Recurrent Unit (GRU) based model for classifying emotions into positive and negative categories. Among these three models BanglaBERT outperformed by F1-score of 0.92, succeeded by LSTM at 0.9049 and GRU at 0.8950, as per our k-fold cross-validation analysis. These findings show how effectively transformer-based models are doing for sentiment analysis in Bangla. For companies and researchers looking to improve natural language processing (NLP) applications in Bangla e-commerce, the findings offer insightful information. Future research will concentrate on investigating multi-class sentiment categorization, increasing classification accuracy, and growing the dataset.