Predicting Organizational Performance: Application of Neurocomputing as an Alternative to Statistical Regression
Abstract
This thesis presents an application of neurocomputing as an alternative function approximation tool for predicting organizational performance. Organizational performance assessment typically involves association of component organizational attributes with a composite ordinal rating. This is a complex generalization problem which is difficult to solve by analytical means. Conventional statistical methods are particularly unwieldy for this application because of the unknown functional domain and multi-dimensional interactions among predictor variables. The back-propagation neural network is an alternative function approximation tool which has been shown to outperform statistical regression in applications lacking a well defined domain model. This research applies this technology to the problem of predicting organizational performance. The content and ratings of Air Force civil engineering unit effectiveness inspection reports were used to train back-propagation neural networks to discriminate between excellent and satisfactory squadrons. These trained networks significantly outperformed logistic regression models in correctly predicting squadron performance ratings during cross-validation trials (81 percent versus 61 percent). Keywords: Neural nets, Cybernetics, Mathematical prediction.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1989
- Accession Number
- ADA216130
Entities
People
- Franklin W. Baugh
Organizations
- Air Force Institute of Technology