Sentiment classification, a critical task for analyzing textual emotions, faces challenges due to the variability in language across different domains. This research introduces a multi-domain adaptive deep learning approach to enhance sentiment classification. The model integrates XLNet and Glove embedding layers to achieve high performance across diverse domains from the Amazon Product Reviews dataset. This approach offers a robust solution for sentiment classification, addressing the limitations of domain-specific models and providing a versatile tool for various applications. Experimental results demonstrate significant improvements over traditional and deep learning models from various baseline studies, with accuracy reaching 91%. The paper introduces a novel multi-domain adaptive deep learning model for sentiment classification, addressing challenges in generalization across different domains. Future research directions include exploring advanced domain adaptation methods, enhancing embeddings, and optimizing training processes for better performance in low-processing-power environments.