A Hybrid Feature Selection Model for Emotion Recognition using Shuffled Frog Leaping Algorithm (SFLA)-Incremental Wrapper- Based Subset Feature Selection (IWSS) Sri Raman Kothuri* Assistant Professor, Department of Computer Science and Engineering, Vel Tech Ranagrajan Dr Sagunthala R&D Institute of Science and Technology, Avadi, Chennai, 600062, India sriramankothuri@veltech.edu.in Dr N R Rajalakshmi Professor, Department of Computer Science and Engineering, Vel Tech Ranagrajan Dr Sagunthala R&D Institute of Science and Technology, Avadi, Chennai, 600062, India drnrrajalakshmi@veltech.edu.in Abstract Emotion recognition method is required for therapy to recognize the emotions of patient and helps in treatment. Many computer science based emotion recognition works focused on facial expression, speech, body gesture and multi-modal based machine learning techniques. Existing methods have limitations of poor convergence and easily trap into local optima. In this research, the Shuffled Frog Leaping Algorithm (SFLA)- Incremental Wrapper-based Subset Selection (IWSS) hybrid method is proposed to improve the emotion recognition. The proposed method involves in analysis the emotion of user through video, audio, and text features and recommends the music to the users. The analysis shows that hybrid modality shows the higher performance in emotion recognition. AlexNet model is applied for the feature extraction in video data and Latent Dirichlet Allocation (LDA) is applied for text feature extraction. Multi-Class Support Vector Machine (MC-SVM) model is used for the classification. The proposed SFLA-IWSS method has 97.05 % accuracy and existing gSpan method has 90 % accuracy. Keywords: AlexNet; Incremental Wrapper-based Subset Selection; Latent Dirichlet Allocation; Multi- Class Support Vector Machine; Shuffled Frog Leaping Algorithm. 1. Introduction In the field of human-computer interaction and artificial intelligence, emotion recognition plays a promising role. Various techniques like heartbeat, blood pressure, body movements, speech recognition, facial expressions and textual information were used to detect emotions of the users (Batbaatar et al., 2019). Individual’s mental state related with behavior, feelings, thoughts are often defined as an emotion. Emotion recognition is one of the popular research in Artificial Intelligent and its ability to mine opinions in social media data such as Twitter, Reddit, YouTube, and Facebook, and others (Poria et al., 2019). Speech is considered as natural way to express ourselves and this is used for emotion recognition. Text is used to way of communication in emails, messages and this is used to recognize the importance of the emotion. Speech Emotion Recognition (SER) is often used for the emotion recognition [Akcay et al. (2020)]. Emotion recognition embedded in a healthcare system to monitor the patient physical and mental state and prescribe suitable medicine or therapy [Hossain et al. (2019)]. Another important module in emotion recognition is facial expression. Facial expression in video is applied to extract the facial features for emotion recognition [Jain et al. (2019)]. Multi-modal emotion recognition is interesting field of research for effective performance of sentiment analysis and computing process. Emotion recognition system is more accurate for different nature of signal carried out for exploiting the information (Nemati et al., 2019). Existing methods treated the features at each time step as e-ISSN : 0976-5166 p-ISSN : 2231-3850 Sri Raman Kothuri et al. / Indian Journal of Computer Science and Engineering (IJCSE) DOI : 10.21817/indjcse/2022/v13i2/221302040 Vol. 13 No. 2 Mar-Apr 2022 354