Stress emotion classification using optimized convolutional neural network for online transfer learning dataset. Issue 14 (26th October 2022)

• Record Type: Journal Article
• Authors: Linda Rose, G.; Punithavalli, M.
• Is Part Of: Computer methods in biomechanics and biomedical engineering. Volume 25:Issue 14(2022)
• Page Start: 1576
• Page End: 1587
• Publication Date: 2022-10-26
• Subjects: stress-emotion domain -- OCNNTL -- O2CNNTL -- OCNN classifier -- regularization learning technique -- concept-drifting data streams; Biomechanics -- Data processing -- Periodicals; Biomedical engineering -- Periodicals; Biomechanics -- Periodicals; Biomedical Engineering -- methods -- Periodicals; Computing Methodologies -- Periodicals; 612.7
• DOI: 10.1080/10255842.2021.2024169
• Languages: English
• ISSNs: 1025-5842

Abstract

Abstract: Nowadays, deep learning methods with transfer learning (TL) makes ease of stress emotion classification tasks. Amongst, an optimized convolutional neural network with TL (OCNNTL) executes OCNN-based classification on emotion and stress data domains to learn high-level features at the top layers. However, it fails to handle the abrupt concept drift in real-time; besides, it end up with huge time complication while on gathering the required data and its transformation. To tackle the aforementioned concerns, a novel online OCNNTL (O2CNNTL) model is proposed; whereas, OCNNTL process initiates in the stress-emotion domain via the prior knowledge acquired by learning the training data both from the stress as well as the emotion domains. Moreover in O2CNNTL model, the concept-drifting data streams are taken into account for solving the online classification by the OCNN classifier; whereas, to enhance the learning efficiency a regularization learning technique is instigated on varied feature spaces. Thus, the proposed O2CNNTL achieves higher efficiency than the state-of-the-art models.