TELF-PSPOC: Three-layer Ensemble Learning Framework for Predicting Student Performance of Online Courses

Virtual learning environment(VLE)MOOC provides large-scale data of resources,activities,and interactions within a course structure for predicting student performance.But it is challenging to extract and learn efficient features from student behaviors.In this paper,a three-layer ensemble learning framework for predicting student performance of online courses(TELF-PSPOC)at an early phase is proposed to analyze data collected from Open University Learning Analytics Dataset(OULAD).First,feature augmentation of student behavior is proposed to enrich current features of student performance,including pass rate and grades of all staged tests,daily clicks of online resources.Second,three-layer ensemble feature learning with heterogeneous classifiers(TEFL-HC)is proposed to benefit the integration of tree model and neural network.Compared with current two-layer ensemble learning,pretraining of features prevents overfitting while using nonlinear regression.The experiment shows that our TELF-PSPOC performs better than several baseline models.Besides,the relationship of the learning results and student behavior via VLE is further discovered.