An Application of Machine Learning in Public Policy: Early Warning Prediction of School Dropout in the Chilean Public Education System

Abstract

2021AbstractSchool dropout is a serious problem worldwide, and contributes to a great deal of poverty and misery. People who have not finished school obviously suffer the consequences, but these extend to all of society since they become a burden due to lack of education and skills for the workplace. Much like poverty, school dropout is complex and multidimensional. Hence, early warning systems that predict which children are at risk of dropping out of school are of the utmost importance, and furthermore, the interventions to rescue these children must be bespoke, i.e., tailored to the specific situation of each child. Much work has been done using traditional methods such as attendance thresholds and logistic regression. However, school dropout prediction by means of applying machine learning is relatively new. In addition, an application that has worked in one country does not necessarily work in another, since the available data sets are different. Therefore, the following question arises: does machine learning enable a more accurate early warning of school dropout specifically in Chile? In this paper we answer this question, testing and comparing machine learning predictive models with a traditional logistic regression, using public databases from the Chilean Ministry of Education. In addition, we offer some practical recommendations for other researchers and policy makers who endeavour to implement practical working early warning systems for school dropout.

Keywords

• Machine learning
• school dropout
• predictive model
• neural network
• decision tree