A Deep Residual Auto-Encoding Approach with Metaheuristic Optimization for Educational Quality Improvement

Abstract

The technology that deals with analyzing the educational data is in vogue in the modern educational setting to help to determine the students’ behavior and learning process. The prediction of students' learning outcomes is essential because it assists the educators in situations in such ways as; early identification of learners who are performing poorly, formulation of the necessary approaches to address the matter, and therefore improving the learners’ performances. Therefore, despite the development of many concepts in this field, some of them have vagueness in most of the results, low compatibility with complicated data structures, or poor quality of other aspects such as optimization that makes predictions less accurate.  Hence, the proposed work develops a smart framework, AcadPredictor that employs advanced deep learning and optimization models for the prediction of students’ academic performance. In the proposed system, the Regression Auto-Encoding Residual Network (Hyb-RARNet) model based on regression, auto-encoder and residual model is developed, which allows the model to distinguish as well as detect the sample data pattern and optimize the EDM. In order to further improve the prediction performance of classifier, the Honey Levy Explorer Optimization (HLEO) is used in this study for optimally computing the learning rate, which significantly enhances the speed and accuracy. Moreover, the Kaggle open source students dataset has been used to test and examine the outcomes of the proposed system using a range of evaluation measures