Aggregating expert predictions in a networked environment

This paper describes an approach for combining the classifications or predi ctions of n local experts into a single composite prediction. We describe a Java-based application that allows a user to Select UP to n prediction exp erts that provide information for assigning an object to one of two predete rmined groups. An advantage of this type of application is that it is capab le of interacting with the Internet in a relatively seamless way. We examin e the accuracy and robustness of our technique by comparing the classificat ion accuracy of our technique,, a maximum entropy-based aggregation techniq ue, and four classification methods on a real-world, two-group data-set con cerned with bank failure prediction. The classificaiton methods studied in this work include Quinlan's C4.5 decision-tree classifier, logistic regress ion, mahalanobis distance measures, and a neural network classifier. Our mo del includes a fundamental component (i.e., a transaction manager) that hel ps improve the general performance of applications that perform network-bas ed classification. This component is found to provide reliable and secure c onnections along with ways to direct traffic across the Internet. Our resul ts suggest three major contributions: (1) a transaction manager increases t he flexibility of a network-based classifier since it is capable of transac ting with one or more specific types of prediction expert(s) over the Inter net; (2) our approach tends to be more accurate than the individual classif ication methods we examined; and, (3) our approach can outperform a recentl y introduced statistically based aggregation technique.