In this study, we used a dataset from the Brazilian school census provided by the Ministry of Education to identify relevant attributes for forecasting the number of students enrolled in a school. This dataset contains 340 characteristic attributes of schools and their respective teaching stages. The large quantity and nature of this data make data analysis more complex, which requires an appropriate method for feature selection to enrollment predictive models. In this sense, this study explores the application of Machine Learning algorithms as a solution to the problem of predicting enrollment, including random forest, multilayer perceptron, linear regression, and support vector regression. We assessed the models’ performance using cross-validation, calculating the MAE, MSE, and RMSE metrics and the algorithms’ execution time. The results revealed that the Spearman correlation method with thresholds of 0.6 and 0.65 can reduce the dimensionality of the data and the execution time of the predictive models