A Neural Network Approach to Pattern Recognition in Marketing Data Analysis: An Analysis of Both the Positive and Negative Aspects of System Performance

Abstract

Neural network research in artificial intelligence, although reported mainly in computer science literature, is being pursued by scientists in a variety of disciplines throughout the world (Hopfield 1987). A neural network is an artificial intelligence approach to computing that is characterized by interconnected, parallel functioning neurons with an operating design philosophy based on the biological brain. With newly available software, this type of network can be trained to accept independent variable values as input, and by determining its own heuristic for their relationship, estimate dependent variable values as output. Common terminology used to indicate this type of neural network processing include: adaptive systems, parallel distributed processors, neurocomputers, and natural intelligence (Stanley 1989). Particularly, where prediction is the main objective, neural analysis is finding marketing applications in place of, or in addition to, multivariate techniques such as multiple regression, discriminant analysis, and logistic regression. This is not to suggest that statistical analysis is in danger of becoming obsolete. Neural analysis can only be expected to provide limited insight into data, and will only be appropriate in a small subset of applications.