Abstract
Student course performance is influenced by several factors including past student grades, previous assessments and past studied courses etc. Predicting the university students’ dropout factor has always been regarded as major concern for MOOC platform providers but requires careful selection of important features. Artificial Intelligence (AI) technology such as Deep Learning (DL) models’ techniques use automatic feature selection techniques, rather than considering the handcrafted based on the most influential learner’s academic, temporal and behavioral features extraction that effects on the model prediction of certified and not-certified rate. To solve this problem, this study proposed the DL-based Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) model for training of university students’ MOOC dataset. The proposed hybrid CNN-LSTM model named CLA-GP captured the spatial features and temporal dependencies from time-series data by removing redundant features using autoencoder learning, while Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) algorithms selected the best subset of features from the autoencoder’s key pattern features space. Extensive experiments conducted on the publicly available Massachusetts Institute of Technology (MIT) and Harvard University MOOC dataset showed that the proposed CLA-GP showed highest accuracy, which is 5% to 10% improvement over existing DL models using different feature selection techniques.
Presenters
Ken Nee CheeSenior Lecturer, Department of Computing and Digital Technology, Faculty of Computing and Meta-Technology, Universiti Pendidikan Sultan Idris, Perak, Malaysia
Details
Presentation Type
Paper Presentation in a Themed Session
Theme
KEYWORDS
Massive Open Online Course, Convolutional Neural Network, Long Short-Term Memory
