Beyond The Gradebook: Machine Learning and LMS Data for True Student Performance

Abstract

Analyzing student’s behavior for performance prediction involves examining various data points and indicators to predict academic outcomes. This research uses the learner’s activity tracker tool collected dataset, which contains three different sets of features including demographic, academic background, and behavior features. This approach combines these data to predict students’ performance based on behavior, providing insights that can help in creating personalized learning experiences. A hybrid ML model is developed to improve the student’s performance prediction. The hybrid ML model combines the ensemble feature selector with Optimized with support vector machine (SVM) classifier. The Linear Neural network (NN) model is used to form a three-layered feature extraction (FE) model. This approach uses ensemble feature selector to select the influential features from the Linear NN extracted features map. Finally, an optimized SVM classifier is developed to predict the students’ performance. The optimized SVM model uses the GridSearchCV method to optimize the regularization parameter (‘C’-value) and kernel options of the SVM model to improve the prediction performance. The performance evaluation analysis shows that the LNN-Ensemble-Optimized SVM based students’ performance prediction approach achieves higher accuracy (98.12%), precision (98.51%), recall (99.23%), f-score (98.1%) rate than comparison approaches on LMS data.