Prediction of retention at historically black college/university using artificial neural networks

摘要

Identification of students at risk is crucial part of early intervention strategies that can lead to improvement of first year retention rates and increase of students' success at a historically black college or university. We investigate the application of multilayer perceptron to provide prediction of students' retention based on attributes available at the students' enrollment and after the first semester. The accurate classification is possible when a proper number of hidden nodes is selected, using a set of variables that include students' major, the amount of unmet financial needs, father's and mother's education level, the results of standardized tests (SAT) and students' success at high school. The classification accuracy significantly increases after the first semester data (number of credit hours passed and GPA) become available.