An application of unsupervised neural networks and fuzzy clustering in the identification of structure in personality data.

摘要

Questions related to the number and definition of broad personality types have long been debated in the personality literature. The present research applied new methods to these old questions and tested the utility of these methods in three large samples with a combined sample size of over 3,000. Thus, replicability and generalizability were emphasized throughout this study, so as to move work on personality typologies from the theoretical realm to a level that admits to stringent empirical tests.; The method developed in this study addressed typological issues raised in previous work, especially the reliable identification of the number of distinct types, the definition of type membership in nondiscrete, "fuzzy" terms, and the identification of nonlinear patterns in the data. The method borrowed heavily from techniques developed in the discipline of pattern recognition and used Big Five scale scores as input variables. The input variables were analyzed with both neural network and fuzzy clustering implementations of the c-means clustering algorithm.; The method was used to derive a personality typology consisting of five broad person types which were identified in six data sets that were collected from three populations and used two personality inventories. The five cluster structure was identified by repeated neural network analyses of the six data sets. The cluster centers or mean personality profiles for each of the five clusters or personality types were found using fuzzy clustering analyses. Additionally, a fuzzy membership value was computed for each case in each of the five clusters. Each case was then assigned to the cluster in which it had the highest membership value.; Of the five personality types identified in this study, three were similar to personality types reported in previous work. The identification of the two previously undetected personality types was attributed to the application of a method that used Big Five scales as input variables, reliably identified an optimal number of clusters and represented the nonlinear patterns in the data.