Using neural network technology for predicting military attrition

Summary form only given, as follows. The United States Naval Academy uses multiple linear regression to predict whether or not an applicant is likely to attrite before completing a four-year course of study. This prediction problem is one of a class of problems in which the relationship between the predictors and the criterion is probabilistic. The study presented explored how neural network technology would compare to regression in problem of this type. When to terminate training in the probabilistic situation is one of the primary questions addressed. A double crossed-validation design was proposed to deal with this problem. Four different neural networks were evaluated, all of which led to better predictive efficacy than linear regression