Deep SVM-Driven Predictive Analytics for Improved Decision-Making in E-Learning

This study employs a dataset acquired from an Algorithm Introductory Class at a Brazilian university, where students were evaluated on "21st Century Skills" and engaged in an online environment allowing social interactions via postings and comments. Extensively generating pertinent inferences from this heterogeneous dataset—which consists of qualitative reactions to student postings as well as quantitative skill assessments—is challenging. Particularly in an interactive, online learning setting, conventional grading methods might not fairly represent a student's aptitudes range. Therefore, a good model is required to investigate these complex interactions and project student development. makes advantage of a Deep SVM architecture by combining Residual Deep Belief Networks (DBNs) for feature extraction with Kernel-based Support Vector Machines (KSVM) for classification. This work provides a novel approach combining Residual Deep Belief Network (DBN) for feature extraction with Kernel Support Vector Machine (KSVM) for classification. Application of residual DBN for feature extraction offers a complete mechanism to control and assess the high-dimensional data generated by grading assessments and online activity of students. The study combines many data sources—online interactions and traditional grades—into a logical framework. For the training dataset, the proposed KSVM achieves an accuracy of 91.4%, which is higher than e-LION (85.3%), OLAP-DGCNN (88.7%), SEt-VDCNN (90.2%), and BR2-2 T-MICE (84.9%).