Sentiment analysis has gained significant attention in natural language processing (NLP) due to its applications in social media monitoring, customer feedback analysis, and opinion mining. However, a major chall enge in sentiment classification is class imbalance, where certain sentiment categories are underrepresented, leading to biased models and reduced accuracy. This research addresses the issue by leveraging data augmentation techniques, specifically synonym replacement and back-translation, to balance the dataset and enhance model performance. Unlike conventional deep learning approaches that rely on complex architectures, this study proposes a simplified yet more effective model by utilizing a Decision-based Recurrent Neural Network (D-RNN) trained on an augmented dataset. Additionally, Aspect-based and Priority-based augmentation techniques are introduced to ensure semantic consistency and emphasize critical contextual information during augmentation. Experimental results demonstrate that the proposed approach effectively reduces class skewness and improves sentiment classification accuracy, surpassing traditional models in performance while maintaining lower computational complexity. This research highlights the significance of data augmentation as a powerful strategy to enhance sentiment analysis, offering a cost-effective and scalable solution without the need for more complex deep learning models.