An Early Student Performance Prediction Using GRU Model

Abstract

The goal of Learning analytics is to determine a student’s performance class over time. Because of distance constraints, it might be challenging to assess how effectively students perform in virtual learning settings, even while it helps teachers to intervene in a timely manner. Many studies created prediction models with data from Massive Open Online Courses (MOOCs). Sadly, the sole purpose for these models was to group students into binary classes according to the courses they had taken. The paper proposes a daily multiclass model to forecast students’ performance using Recurrent Neural Networks (RNN), particularly Gated Recurrent Units (GRU), in an effort to close this gap. In this instance, the GRU model was evaluated against two baseline models, the artificial Neural Network and Long Short-Term Memory, or ANNLSTM, in order to determine the model’s validity. GRU outperforms the ANN_LSTM model, which achieved 88% accuracy, with an accuracy of almost 90%. This illustrates how the architecture and ability of the GRU time series model to maintain latent dependencies can be used to predict student success in MOOCs at an early stage.