Predicting Computer Science Student's Performance using Logistic Regression

Abstract

The prediction of student academic performance has drawn noticeable attention in higher education institutions. It may help educational institutions in providing quick assistance to students who are low performed at an early stage. However, it becomes challenging to predict the performance of students because the numerous amount and complexity of the variables such as student's demographics information, student's achievement data, and student's activity data. The purposes of this study are to develop a predictive model using Logistic Regression algorithm on student data coming from a computer science faculty in an Indonesian private university and to identify variables in playing a part in student's performance. The dataset is gathered from the past decade's student records: 2010–2020. It is then explored and preprocessed using various approaches such as merging, aggregating, feature encoding, and SMOTE. About 75% of the dataset is used for training and the remaining is for the testing set. Out of 20 variables used, 10 variables are statistically significant to the response variable. The results show that Logistic Regression achieves a considerably good predictive model with an accuracy rate of 91%, recall rate of 98%, and precision rate of 85%. The model can be used as an early warning system to identify low-performing students and inform both the faculty members and the students. Faculty members can then employ a range of strategies to communicate with low-performing students and give them with opportunities to improve their performance.