Sentiment analysis is a common approach to measuring customer satisfaction, which plays a critical role in customer retention. While traditional approaches rely solely on textual reviews to predict customer sentiment, in this paper, a novel framework is proposed to expand on the traditional approaches, by adding additional data to textual data. The developed fusion framework uses decision-level fusion architecture to combine predictions from BERT (Bidirectional Encoder Representations from Transformers) and Extreme Gradient Boosting (XGBoost). The E-Commerce Clothing Reviews dataset was chosen to evaluate the proposed fusion approach, as it is a real use-case and for having both textual and numerical data. To establish a baseline, individual assessments were carried out for each data type: BERT was trained on textual data, while XGBoost was trained on numerical data. The results demonstrate the promising performance of the decision-level fusion framework, outperforming BERT with a macro-average F1 score of 86.65%.