The growth of e-commerce platforms has led to a huge amount of data, providing useful insights into consumer behavior and preferences. However, the quality of this data is often affected by discrepancies, inaccuracies, and noise, making reliable sentiment and emotion analysis difficult. This paper proposes a novel approach to improve sentiment and emotion analysis in Bangla E-commerce dataset by first identifying and mitigating discrepancies. Using a robust approach these inconsistencies are thoroughly identified and resolved. Next, we apply state of the art model BERT for sentiment analysis and emotion recognition on the cleaned dataset, extracting sentiment polarity and emotion categories from text data related to Bangla e-commerce transactions. The experimental results show significant improvements in sentiment and emotion analysis accuracy after removing dis- crepancies. Initially, sentiment analysis achieved an accuracy of 93.96%, but this increased to 96.31% after addressing the discrepancies. Similarly, emotion analysis accuracy improved from 90.94% to 93.59% after cleaning the dataset. By conducting comparative analyses with traditional methods, we demonstrate the effectiveness of our approach in improving insights from e-commerce data. Our study highlights the importance of discrepancy detection as a key preprocessing step in e-commerce data analysis. By correcting data inconsistencies before sentiment and emotion analysis, we enable a more accurate understanding of consumer sentiments and emotions.