Sentiment Analysis on Twitter Using DSA-Inspired Feature Selection and Multi-Model Comparison

Social media generates massive amounts of textual data, making sentiment analysis crucialfor understanding public opinion. This work presents a simple yet effective approach for sentimentanalysis on the Sentiment140 dataset, introducing a DSA-inspired preprocessing step using a stack toretain the most relevant words, combined with lemmatization and TF-IDF vectorization. We evaluateseven machine learning models — Logistic Regression, Naive Bayes, Linear SVM, Random Forest,Gradient Boosting, Decision Tree, and K-Nearest Neighbors — and compare their performance usingaccuracy, precision, recall, and F1-score metrics. Experimental results demonstrate that LogisticRegression achieves the highest accuracy of 0.7597, followed closely by Linear SVM (0.7521). Theproposed pipeline, accompanied by visualizations such as word clouds, confusion matrices, and modelcomparison charts, provides a comprehensive framework for sentiment analysis on large-scale textualdatasets