Deep Learning for Predicting Students' Academic Performance

Abstract

The use of Deep Learning to predict what happens in the future becomes more popular because of great availability of data. This study proposes the power of Deep Learning to predict Students' Academic Performance (SAP). It is happened, students pass or fail at the end of the school period. By the time, it is already too late to help the students. The prediction of SAP in advance is needed to prevent students from failing. Feedforward Neural Network is used in this study to predict whether students can achieve certain scores within a specific range score. Data from students' activities are used to predict SAP. Raw data were collected from a course attended by 90 students. The input for this study was student's activities that were recorded in the log data from the Learning Management System. The way the data were collected, pre-processed and developed are explained. The result shows that students' activity from log data varied greatly and the formed pattern is weak. The training validation model's accuracy was still around 85% with the accuracy to predict is under 50%. Thus, our initial research results indicate that Deep Learning does not perform high accuracy for predicting Students' Academic Performance because of several reasons, among others, small sample of dataset, relevance of students' activities, and so forth. Regardless of that, the study reveals that with our dataset structure of students' activities, SAP can be predicted. Further study is to investigate more relevance students' activities, more dataset of students, and a longer period of study time.