Learning Performance of International Students and Students with Disabilities: Early Prediction and Feature Selection through Educational Data Mining

Abstract

The learning performance of international students and students with disabilities has increasingly attracted many theoretical and practical researchers. However, previous studies used questionnaires, surveys, and/or interviews to investigate factors affecting students’ learning performance. These methods cannot help universities to provide on-time support to excellent and poor students. Thus, this study utilized Multilayer Perceptron (MLP), Support Vector Machine (SVM), Random Forest (RF), and Decision Tree (DT) algorithms to build prediction models for the academic performance of international students, students with disabilities, and local students based on students’ admission profiles and their first-semester Grade Point Average results. The real samples included 4036 freshmen of a Taiwanese technical and vocational university. The experimental results showed that for international students, three models: SVM (100%), MLP (100%), and DT (100%) were significantly superior to RF (96.6%); for students with disabilities, SVM (100%) outperformed RF (98.0%), MLP (96.0%), and DT (94.0%); for local students, RF (98.6%) outperformed DT (95.2%) MLP (94.9%), and SVM (91.9%). The most important features were [numbers of required credits], [main source of living expenses], [department], [father occupations], [mother occupations], [numbers of elective credits], [parent average income per month], and [father education]. The outcomes of this study may assist academic communities in proposing preventive measures at the early stages to attract more international students and enhance school competitive advantages.