Truxton Fulton, Simon Kasif, Steven Salzberg, David Waltz
Decision trees are an important data mining tool with many applications. Like many classification techniques, decision trees process the entire data base in order to produce a generalization of the data that can be used subsequently for classification. Large, complex data bases are not always amenable to such a global approach to generalization. This paper explores several methods for extracting data that is local to a query point, and then using the local data to build generalizations. These adaptively constructed neighborhoods can provide additional information about the query point. Three new algorithms are presented, and experiments using these algorithms are described.