Data fusion by intelligent classifier combination

The results of different classifiers in a hybrid intelligent system were discussed for dealing with problems of data mining in the pharmaceutical industry. It was shown that genetic programming is a useful tool for evolving appropriate combination functions when individual classifiers are tunable over a range of decision thresholds and are capable of delivering a confidence estimate with their output. Care was taken to avoid the use of over fitting both in producing the individual classifiers and in design of the genetic program. The technique was used by producing combined classifiers for a publicly available Landsat example and for a pharmaceutical problem.