Estimating Models with Binary Dependent Variables: A Neural Network Approach

摘要

Recently there has been a considerable interest in artificial neural network models that learn to produce 'outputs' from repeated experiences with a set of input and output pairs [8] [9] [26] [27] [41]. These network models achieve computational power via dense interconnection of simple processing elements. Neural net models have the greatest potential in applications requiring the simultaneous parallel operation of many processing elements. One such area of application is the classification problem in discriminant analysis. It involves the development of a decision rule to assign entities (individuals or observations) having q (q>=l) characteristics to one of the two or more given groups. The classification criterion is derived using the historical sample data for which the group memberships are already known and the resulting criterion is then used to place unclassified entities into one or the many prespecified groups.