A Multi-Outcome Hybrid Model for Predicting Student Success in Engineering

In this work, we propose a backpropagation neural network model to predict retention and college GPA of engineering students simultaneously. Unlike previous models, which can only predict single outcomes, this method is capable of modeling two outcomes in the same model. A multi-outcome model and two single-outcome models are developed and tested on 1470 firstyear engineering students who enrolled in a large Midwestern university during the 2004-2005 academic year. The predictors of the models include seven affective measure factors and eleven high school history factors. The affective measure factors are leadership, deep learning, surface learning, motivation, meta-cognition, expectancy-value, and major decision. The high school history factors are high school GPAs, standardized test scores, and the grades and number of semesters in math, science, and English courses in high school. In the multi-outcome model, the overall accuracy of retention prediction is 71.3%. The mean error of GPA prediction is 16.5% of the full scale. High school grades and SAT scores are better predictors of college GPA, while the number of semesters of English, science, and math in high school and affective measure factors (motivation, leadership, etc.) are better predictors of first-year retention. This work lays a foundation for modeling multiple success outcomes in the future.