Fusing statistical and neural classification for screening undergraduate pilot training candidates

Every year the US Air Force spends millions of dollars to send personnel through Undergraduate Pilot Training (UPT) and other training programs. Identifying the most qualified candidates is a difficult, yet critical task. This study applies multivariate data analysis techniques, including discriminant analysis and neural networks, to develop a model to predict candidate success during UPT. An entire cradle to grave approach is presented from data screening to model implementation. The model is validated to establish its predictive accuracy, capabilities, and limits. The overall study demonstrates the power of fusing statistical and neural classification techniques for increasing the power of predictive models.