A Comparative Analysis of Machine Learning Techniques for Emotion Detection: A Performance Evaluation Study

This paper presents an emotion-based method for evaluating people's dynamics in interpersonal contexts using textual tweets data and facial expressions. The proposed model sorts facial expressions into groups called "Anger," "Sadness," "Fear," and "Joy." A study of the literature is done to find new writings on decision-level fusion and multimodal emotion detection. Text and facial multimodal expression recognition techniques were used together to collect data for this study. This gave researchers more accurate and nuanced information about how people interact with each other. This method can be used to do things like check social media, look at customer comments, and check on mental health. Using both text and face traits together, group decision classification can figure out how someone is feeling by looking at the message. For extracting face features and text features using the BagofWords method, a convolutional neural network is used. The Group decision classification method was chosen as the probability classifier for natural language processing because it did the best job on the training dataset. The group decision classification and decision-level fusion methods are tested using statistical analysis to see how well they work. With an accuracy of 97% and an f--measure of 94%, the suggested work does a better job. We suggest a decision-level fusion method for identifying emotions across multiple modalities, which is seen as a late fusion in which fine-tuned models are made for each modality.