Learning analytics with correlation-based SAN-LSTM mechanism for formative evaluation and improved online learning

Abstract

Learning analytics (LA) is the measuring, gathering, analyzing, and reporting of data on learners and their environments. This data is used to analyze and improve e-learning. LA has traditionally been used for a variety of purposes, including the prediction of student academic progress and more specifically, the identification of students who are in danger of failing a course or quitting their studies. However, the majority of the existing schemes have the issue of accurately predicting the students' performance in online courses. In this paper, correlation-based self-attention networklong short-term memory (SAN-LSTM) is proposed to predict students' outcomes, along with the effectiveness of the teaching experience, as well as the assessment methods. Initially, the data is collected from three datasets namely, WorldUC, Liru, and Junyi to evaluate the performance of the proposed approach. The min-max normalization is employed to improve the performance of the approach. The correlation-based feature selection (CFS) is employed to select appropriate features from the pre-processed data. Finally, the correlation-based SAN-LSTM is established to forecast the effectiveness of fine-grained learning. Three real-world datasets gathered from various e-learning empirically validated that the proposed model improves prediction outcomes and provides useful data for formative evaluation. The existing methods such as adaptive sparse self-attention network (AS-SAN), Bangor engagement metric (BEM), and deep belief network learning style (DBNLS) are used for comparison to justify the effectiveness of the correlation-based SAN-LSTM method. The proposed correlation-based SAN-LSTM achieves better results of 98% of accuracy and 93% of precision. The proposed method achieves 98% accuracy which is higher when compared to those of AS-SAN, BEM, and DBNLS.