Comparison of machine learning algorithms for censored regression models of University scores

Abstract

This study compares 12 machine learning algorithms for predicting university rankings. We have collected scores of 316 universities on QS World University Rankings across 9 countries namely Australia, Brazil, United States, India, Germany, France, China, Japan, and Russia. In this data, only values of top universities with scores above a certain are available. This means that for the other universities, the scores are censored. These scores are predicted from data that are publicly available about the universities; examples are the number of publications or the total number of students. Cross-validation with the concordance index is used for model comparison. It is found that gradient boosting machines and random survival forests perform best on this application.