Multivariate Sequential Modelling for Student Performance and Graduation Prediction

Abstract

High dropout rate and low student performance were inevitable issues for educational institutions in many countries. Consequently, this study presents an automated technique to predict student performance and graduation using student data with separated and combined prediction method. The data was collected from an Indonesia university. Long Short-Term Memory (LSTM) and Gate Recurrent Units (GRU) as an outstanding model in handling sequence data was proposed in this study. According to our study, both LSTM and GRU have a great performance above 90% in predicting each task. The performance of both architecture was surpass each other depending on the corresponding task. In early prediction, the student graduation prediction can give a satisfiable performance since the first semester, despite having tradeoff in recall. Whereas in student performance prediction, the RMSE value was acceptable since the second semester. Overall, the performance of student performance and graduation prediction was better if used separated method than combine method. This pipeline work can be replicate and improve for similar task in other universities with feature adjustments based on data availability.