Predicting Dropout in Higher Education Based on Secondary School Performance

Predicting student performance, preventing failure and identifying the factors influencing student dropout are issues that have attracted a great deal of research interest recently. In this study, we employ and evaluate several machine learning algorithms to identify students at-risk and predict student dropout of university programs based on the data available at the time of enrollment (secondary school performance, personal details). We also present a data-driven decision support platform for education directorate and stakeholders. The models are built on data of 15,825 undergraduate students from Budapest University of Technology and Economics enrolled between 2010 and 2017 and finished their undergraduate studies either by graduation or dropping out. We handle the problem of missing data by imputation. After performing feature extraction and feature selection, a wide range of classifiers have been trained including Decision Tree-based algorithms, Naive Bayes, k-NN, Linear Models and Deep Learning with different input settings. The methods were tested using 10-fold cross-validation and the AUC of the best models, Gradient Boosted Trees and Deep Learning, were 0.808 and 0.811 respectively.