Does machine learning risk reinforcing societal prejudice in education?

Abstract

Machine learning is increasingly being applied in sectors ranging from healthcare to finance; however, in education, it is typically only used for predicting students’ grades. On the other hand, deeply rooted societal prejudice is more challenging to measure, so could machine learning contribute to the current discourse? As a result of a gap in existing literature in the use of machine learning in education, this study uses this novel approach to investigate the potential links between the levels of prejudice of college students and their parents’ levels of education. An Implicit Association Task (IAT) was used to collect the information from the participants. Before applying three different machine learning models: Decision Tree (DT), Support Vector Machine (SVM), and K-Nearest Neighbour (KNN). It was found that KNN marginally outperformed not only the DT model but also SVM, with the results being validated by using the Statistical Package for Social Scientists (SPSS). This demonstrated a clear correlation between the parents’ education and their children’s prejudice levels. The paper adds to the limited research that is available on the use of machine learning in education and proposes that a larger study be conducted to provide a more nuanced understanding of prejudice in education.