Application of neural networks to classification of binary profiles derived from individual interviews

A practical application of a trainable neural network has been developed for building classifiers of complex binary patterns. The patterns consist of codes assigned in scoring responses to the question-like stimuli presented in interviews constructed to characterize individual personality classes. In an idealized situation, the testing instrument used for acquiring observational data has been designed to obtain pattern data which support a linear separability of classes of individuals and to exhibit a high degree of independence among the binary variables. With such data, a network with only a single layer of weights is sufficient. When complex, convoluted interactions among the variables play a critical role in the classification, a more complex architecture of weights must be realized by including more than one layer of connections or weights. Experiments were directed at using neural network ideas in deriving sets of weights capable of performing classifications. These simulation experiments spanned a range of complexities for the connecting layers. Experimental results are presented.<>