Machine Learning Prediction Model for Early Student Academic Performance Evaluation in Video-Based Learning

The transformation of education norms from face-to-face teaching era to the Massive Open Online Courses (MOOCs) has created the emergence of educational technology domain for many students to access e-learning platforms. However, there are some drawbacks especially in asynchronous video-based learning. A sense of isolation could occur between teacher and students if the teachers do not interact much with the students in the asynchronous video-based learning. Consequently, the knowledge that is delivered by the teacher may not reach students effectively and cause a drop in student performance in the coming examination. Moreover, the growth of video-based learning has created a huge amount of data on the student learning process on the educational video which may provide a boost for educational data mining research. Therefore, this research study aims to introduce a predictive model that scrutinize the number of video view data based on each chapter in the video as well as student learning style, Felder-Silverman (FS) learning style model to deliver a prediction on individual student early performance in asynchronous video-based learning. This research has tested the different combination of feature selection methods with several handle of imbalance data methods such as Synthetic Minority Oversampling Technique (SMOTE), SMOTE-TOMEK and Adaptive Synthetic (ADASYN) algorithms to build the machine learning model and compare the model performance. As a result, proposed machine learning classifier algorithms with the combination of Maximum Relevance and Minimum Redundancy (MRMR) as feature selection method and SMOTE has been achieved the highest Area Under Curve (AUC) rate of 0.93