Dynamic Integration of Data Mining Methods Using Selection in a Knowledge Discovery Management System

One of the important directions in improvement of the data-mining and knowledge discovery methods is the integration of multiple classification techniques. An integration technique should estimate and then select the most appropriate component classifiers from an ensemble of classifiers. We discuss an advanced dynamic integration technique with multiple classifiers as one variation of the stacked generalization method based on the assumption that each component classifier is the best inside some sub areas of the application domain. In the learning phase a performance matrix of each component classifier is derived and it is then used during the application phase to estimate the performance of each component classifier with new instances.