Classifying Student Success from Their Behavioral Pattern in Online Learning using Machine Learning Approach Rr Taraningrum Puspa Wijaya (1), Ervaran Panjilara Putra (1), Nora Hariadi (1), Devvi Sarwinda (1), Bevina Desjwiandra Handari (1, a)
1) Department of Mathematics, Faculty of Mathematics and Natural Sciences (FMIPA), Universitas Indonesia, Depok 16424, Indonesia
a) bevina[at]sci.ui.ac.id
Abstract
Students academic activity in Learning Management Systems (LMS) is strongly correlated with their academic results in online learning. This behavioral activity is dynamic and difficult to predict. However, machine learning can predict their possible academic, behavioral patterns and classify their future results. This research aims to classify student success through student behavior patterns represented by features on the LMS. The prediction is required to prevent poor student performance. After discovering student behavioral patterns, classification is carried out to determine students success in online learning. The machine learning methods used in this research are Recurrent Neural Network (RNN) and Support Vector Machine (SVM). RNN is used to predict student behavioral patterns through time-series data activity recorded in E-learning Management Systems (EMAS), specifically, LMS in UI. SVM is used to determine if students pass online learning courses or not. Classification in SVM incorporates a pass/fail category by creating an optimal hyperplane to separate categories. This research uses some Recursive Feature Elimination-Random Forest features in RNN to select most academic activity features that affect the online learning process. The implementation uses four data proportions: 60%, 70%, 80% and 90% training data. Based on the data, the best RNN model uses a hyperparameter based on the number of nodes in the input, hidden, and output layers, which are 1, 10, and 1, and a learning rate of 0.01, and 500 epochs on 60% training data. This RNN method predicts three student behavioral patterns (course view, discussion view, and file view), resulting from the selection feature with an MSE score of less than 0.005. By implementing the same academic activity features on SVM, the best classification for the pass category over ten attempts with 60% training data are a mean accuracy, precision, recall, and F1-score of 88.9%, 91.1%, 97.2%, and 93.9%.
