Sentiment classification plays a crucial role in natural language processing (NLP), yet domain dependency remains a significant challenge, limiting model generalization across different domains. This study evaluates the effectiveness of traditional machine learning models—Support Vector Machines (SVM) and Random Forest (RF)—and deep learning architectures—Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM)—for cross-domain sentiment classification. Using TF-IDF representations, models were trained on product reviews from the electronics domain and tested on Books, Beauty & Personal Care, and Automotive. Performance was assessed using Accuracy, F1 score, and AUC, with tenfold cross-validation ensuring robustness. Results indicate that deep learning models outperform traditional approaches, with LSTM achieving the highest classification accuracy, followed by CNN. Among traditional models, Random Forest outperformed both SVM variants, while SVM with an RBF Kernel struggled to generalize. Despite improvements with deep learning, cross-domain effects persist, suggesting the need for enhanced feature representations and domain adaptation techniques. These findings provide insights into model trade-offs, contributing to advancements in cross-domain sentiment classification.