Comprehensive Evaluation of Sentiment Analysis using Machine Learning Techniques

Sentiment analysis, a key subfield of natural language processing (NLP), is essential for extracting and interpreting opinions in text across diverse domains such as e-commerce, social media, and finance. Traditional lexicon-based techniques often fall short in handling contextual ambiguity, which has led to a shift toward machine learning (ML) approaches. This review investigates various sentiment classification models, including supervised, unsupervised, and deep learning techniques such as Multinomial Naïve Bayes, Support Vector Machines (SVM), Random Forests, Decision Trees, and ensemble methods like Voting Classifiers. It also explores deep learning models including Long Short-Term Memory (LSTM) networks and Transformer-based architectures such as BERT and GPT-3. Among the models evaluated, Random Forest achieved the highest F1-score (97.41%), while SVM and Voting Classifier also demonstrated strong performance. However, Transformer models show promise for future improvements despite limited comparative metrics in current studies. Key challenges—such as sarcasm detection, multilingual processing, and bias mitigation—are discussed. The contribution of this review lies in offering a comparative analysis of various ML models based on performance metrics and identifying future directions like multimodal sentiment analysis, domain adaptation, and explainable AI to improve classification accuracy and trustworthiness.