Neural networks as nonlinear models in Air Force personnel analysis: a prospectus and exploratory results

Due to the character of the original source materials and the nature of batch digitization, quality control issues may be present in this document. Please report any quality issues you encounter to digital@library.tamu.edu, referencing the URI of the item.Includes bibliographical references: p. 146-157.Issued also on microfiche from Lange Micrographics.This research evaluates the potential for applying neural networks to Air Force personnel analysis through a review of relevant literature and empirical testing of neural networks in the domain of personnel research. Neural network technology has recently demonstrated capabilities in areas important to personnel research such as statistical analysis, decision modeling, control, and forecasting. An extensive review of the neural network literature indicates that these networks have proven superior to more traditional analytic techniques in many applications. This review also indicates that three different neural network architectures (back propagation, learning vector quantization, and probabilistic neural network) are particularly suited to modeling many aspects of the Air Force personnel system. As demonstrated in the literature, the principal benefit offered by these architectures is the ability to derive nonlinear and interacting relationships among the components of a model. Combined with an examination of current Air Force personnel models, the review of neural network literature indicates several personnel modeling areas which could benefit from the added flexibility of the neural network architectures. In particular, four specific areas are empirically examined using traditional models and neural network techniques: (1) reenlistment modeling and projection, (2) undergraduate pilot training (UPT) selection, (3) aggregate personnel flow-rate projection, and (4) productive capacity analysis. Each of the four areas is examined by comparing the performance of existing models with the performance of neural network models developed using the same information. Special care is takento preclude the effects of over-fitting by making all final comparisons on data not utilized in model development. Overall, the research suggests that neural networks offer the potential for improving modeling in many personnel areas provided care is taken to avoid "over-traininie' to sample data