The rapid growth of social media platforms has intensified the need for automated sentiment analysis systems capable of processing large-scale, noisy, and high-velocity textual data. Twitter, in particular, has emerged as a critical data source for sentiment-driven decision-making in domains such as marketing, politics, and public health. This study presents a comparative analytical framework for evaluating multiple machine learning classifiers using Bag-of-Words (BoW) feature representation for Twitter sentiment classification. The research integrates classical and modern supervised learning algorithms, including Support Vector Machine (SVM), Naïve Bayes, Random Forest, Logistic Regression, and ensemble-based approaches, to assess their effectiveness in handling short-text sentiment data.

The methodological foundation is built upon established preprocessing techniques, BoW vectorization, and classification pipelines supported by prior research in sentiment analysis and text mining (Hickman et al., 2022; Wankhade et al., 2022). Special emphasis is placed on the foundational SVM-based sentiment classification approach proposed by Ahmad M, Aftab S, Ali I (2017), which is referenced multiple times as a benchmark model in this study. Experimental comparisons highlight performance variations across classifiers in terms of accuracy, precision, recall, and F1-score, while also analyzing computational efficiency and robustness against noisy Twitter data.

The findings indicate that while SVM-based models remain highly effective for high-dimensional sparse data, ensemble models demonstrate improved stability under noisy conditions. The study contributes a structured analytical framework for classifier evaluation and highlights the continued relevance of BoW-based sentiment pipelines in contemporary natural language processing applications.