A Study on Learners' Behavior Mining and Prediction to Improve the Learning Performance in Virtual Learning Environment

[[abstract]]Early prediction of student learning performance and analysis of student behavior in a virtual learning environment (VLE) is critical to improving learning performance and minimizing failure rates in online courses especially during the COVID-19 pandemic. However, the predictive performance of traditional machine learning models for at-risk students is poor due to imbalanced data distribution between “pass” and “fail” students in the VLE. Early prediction of learning performance is a relatively new concept in the field because it involves techniques for assessing students as early as possible in their learning intervals to retain at-risk students and provide appropriate procedures and interventions to monitor and minimize failure rates. Learning performance could also be improved by emulating and adopting the learning behavior of students with good learning performance. In this study, we propose early prediction and observing learning behavior to improve student learning performance. The early prediction for the at-risk student model was developed using data resampling to handle an imbalanced dataset and a combination of convolutional neural network (CNN) and long-short-term memory (LSTM) layers. The proposed model achieves higher prediction performance than the baseline ML and LSTM models. Moreover, the new online flipped classroom designed for a hands-on online course gives better results than the previous online course. Clustering and sequence pattern mining were employed to identify good student behavior while participating in the course. A good learning pattern could be used as a recommendation that future learners could emulate to improve their learning performance.[[abstract]]Early prediction of student learning performance and analysis of student behavior in a virtual learning environment (VLE) is critical to improving learning performance and minimizing failure rates in online courses especially during the COVID-19 pandemic. However, the predictive performance of traditional machine learning models for at-risk students is poor due to imbalanced data distribution between “pass” and “fail” students in the VLE. Early prediction of learning performance is a relatively new concept in the field because it involves techniques for assessing students as early as possible in their learning intervals to retain at-risk students and provide appropriate procedures and interventions to monitor and minimize failure rates. Learning performance could also be improved by emulating and adopting the learning behavior of students with good learning performance. In this study, we propose early prediction and observing learning behavior to improve student learning performance. The early prediction for the at-risk student model was developed using data resampling to handle an imbalanced dataset and a combination of convolutional neural network (CNN) and long-short-term memory (LSTM) layers. The proposed model achieves higher prediction performance than the baseline ML and LSTM models. Moreover, the new online flipped classroom designed for a hands-on online course gives better results than the previous online course. Clustering and sequence pattern mining were employed to identify good student behavior while participating in the course. A good learning pattern could be used as a recommendation that future learners could emulate to improve their learning performance