The predictive accuracy of computer-based classification decision techniques. A review and research directions

Computer-based classification decision (CBCD) techniques can be important assets to organizations. However, empirical research evaluating CBCD performance has been inconsistent, resulting in a lack of understanding concerning various techniques' relative merits. An important reason for this is the absence of a theoretically-based research framework that can increase the productivity of CBCD empirical work. Employing statistical prediction theory, this paper provides such a framework. Research productivity can be improved by using the framework to help focus investigations on potentially important areas. A review of the empirical CBCD literature indicates that, though many of the propositions derived from the framework are acknowledged as important, few have been examined. Research productivity can also be improved when researchers are able to make judgments concerning the similarity of research contexts. For example, if characteristics important to CBCD accuracy are used to describe data sets employed, subsequent research can use such descriptions to create research designs that build on and extend previous research efforts. The framework offered in this paper enables such judgments.