Pre-processing using Topographic Mappings

We review two recently developed methods which are used to improve classifier accuracy, bagging and the random subspace method. Both of these methods (and other similar methods) may be characterized as deleting some of the information in the training set and creating classifiers which, though themselves sub-optimal, may be combined to create a better classifier than that created using the original data. We pre-process the data using an unsupervised method which creates topographic mappings and show that the resulting classifiers exhibit diversity and better performance.