Sentiment Analysis (SA) is a widely utilized contextual mining technique for extracting subjective and valuable insights from data. Leveraging Natural Language Processing (NLP), Artificial Intelligence (AI), and computational methods, it identifies, interprets, and analyzes emotions, reactions, or sentiments embedded within data. Feature analysis is pivotal in enhancing and optimizing SA models. With advancements in machine learning (ML), SA has expanded to various modalities and their combinations. In this study, we employ feature extraction and selection techniques for sentiment classification, framing the task as a multi-class classification problem. Using the openly available CMU-MOSI dataset, we assess the performance of multiple ML algorithms using the different feature selection methods. These models are applied to text, audio, and video data individually (unimodal) and in combinations of two (bimodal) and three (trimodal) modalities. Our results demonstrate that the trimodal approach delivers the highest accuracy.