Academic Performance Prediction and Assessment Patterns Among University Students in Technology-Based Courses

Abstract

Predicting and understanding student performance patterns in technology-empowered higher education is significant for predicting student performance and making better educational decisions. The purpose of this study was to analyse the academic performance pattern, explore the relationship of the assessment indicators at a particular course level to cumulative grade point average (GPA) and to compare the performance of the various statistical and machine learning models in predicting academic outcomes. A quantitative, retrospective observational design was used, secondary academic records were obtained from a publicly available repository, Mendeley Data. This study utilized 1,656 undergraduate automotive engineering student records from which the following demographic characteristics, course grades and cumulative GPA were retrieved. Descriptive statistics, Pearson correlation analysis and predictive modeling were used. The performance of four prediction methods was tested namely, Linear Regression, Decision Tree Regression, Random Forest Regression, and Support Vector Regression. Analysis of the results showed that the mean cumulative GPA was 2.40 ± 0.51 with comparatively high GPA in practical training and specialized engineering subjects as compared to the fundamental courses. The technical diagnostics (r = 0.551) and Computer-Aided Design (r = 0.531) were relatively positively correlated with cumulative GPA. Random Forest Regression (R² = 0.517) was the second best predictive model, with the best predictive model being Support Vector Regression (R² = 0.578). Demographic variables were limited in scope and there are no individual assessment components, which limits the interpretation of the results.