Quality of interaction-based predictive model for support of online learning in pandemic situations

Higher education institutions place a lot of importance on their electronic learning systems. Educational institutions in Pakistan and other countries have adopted learning management systems (LMS) due to the coronavirus (COVID-19) pandemic scenario. The learning management system (LMS) establishes a digital learning environment where evaluation and user learning behavior must be carefully analyzed. The “quality of interaction” (QoI) of students is one of the main issues in LMS. Based on various usage matrices (such as the number of logins, clicks, total time spent on the LMS, and actions taken), a student’s level of interaction with the LMS can be determined. QoI is an essential predictor of the accomplishment of students’ final grades. Normally, to examine the effectiveness of LMS usage on students’ learning performance, studies have relied on data gathered from users via surveys. However, the data gathered through surveys are typically associated with the risk of distortion or low quality. Therefore, in order to evaluate and predict the quality of interaction in terms of usage matrices, our proposed work analyzed data from the Moodle LMS at “Hazara University” (HU) for the law and English departments’ courses. This research aims to assess and forecast the quality of student interaction within an LMS by analyzing usage metrics. Unlike traditional survey-based approaches, we explored the predictive performance of LSTM (Long Short-Term Memory), Exponential Smoothing method (ETS), and ARIMA (Autoregressive Integrated Moving Average) methods to predict the weekly LMS usage factors of students. ARIMA and ETS produce better prediction results than LSTM for weekly predictions. Moreover, LSTM model training took considerable computational time for provided datasets.