Educational Data Mining: Predicting Students' Academic Achievement.pdf

This doctoral dissertation explores the application of Educational Data Mining (EDM) techniques to predict students' academic achievement. It provides a comprehensive overview of EDM, focusing on academic prediction tasks, and offers insights into various data mining methods used in education. The study includes a literature review covering relevant works from 2007 to 2022.

Two datasets were analyzed: one from Princess Norah University in Riyadh, Saudi Arabia, containing records of 300 undergraduate students in Computer and Information Science, and another from the University of Mannheim, Germany, with data on over 700 Business Informatics master's students.

Through comparing eight data mining algorithms (C4.5, Simple CART, LADTree, Support Vector Machine, Naïve Bayes, K-nearest-Neighbor, Artificial Neural Networks, Random Forest) using Weka software and 10-fold cross-validation, the research identified key predictors of academic achievement:

- Princess Norah University: Student GPA, failed courses, and grades in core courses are most influential. Naïve Bayes and Random Forest algorithms performed best. English proficiency and orientation year attendance were found to have minimal impact.

- University of Mannheim: Bagging (Random Forest) and Boosting algorithms outperformed individual classifiers. Semesters' grades, student culture, and distance from accommodation to campus were significant features.

The findings can inform the development of recommender systems for timely interventions, benefiting students at both undergraduate and postgraduate levels in relevant fields. Key terms include student performance, machine learning, EDM, student dropout predictions, imbalanced datasets, and oversampling methods.