Forecasting Student Enrollments in Brazilian Schools for Equitable and Efficient Education Resource Allocation

In recent years, there has been growing scientific interest in developing effective techniques for forecasting student enrollment across the school spectrum (i.e., primary, secondary, and higher education). Enrollment forecasting is crucial in shaping public education policies by guiding resource allocation and ensuring equitable access to educational opportunities. In this sense, Machine Learning (ML) models emerge as a promising approach to forecasting the number of students that should be enrolled in a given school term by considering the high complexity of grouping and identifying useful patterns in the prediction process. In this work, we develop a predictive model based on the Random Forest (RF) algorithm to forecast the enrollment of students across the entire spectrum of Brazilian education. We use a database provided by the National Education Development Fund (FNDE), a Brazilian government body responsible for purchasing and distributing textbooks to all public schools. We generate 1,531,185 time series to serve as an input to RF processing. Our training dataset utilized data between 2010 and 2020, and our testing dataset utilizes data from 2021. As a result, RF obtains a higher performance in all the investigated scenarios concerning the Exponential Smoothing (ES) baseline algorithm. Since RF demonstrated acceptable performance, the Brazilian government could benefit from this forecasting technique for student enrollment in school environments and to ensure equitable access to essential resources, such as didactic materials, for the students.