The goal of this project is to improve attribute selection in data preprocessing. This is done using two techniques, attribute combination and clustering. Combination generates new attributes by combining pairs of numeric attributes with arithmetic operations. Attribute clustering discovers groups of categorical attributes based on similarity via Minimum Description Length. The combinations generated frequently have increased correlation to the target attribute compared to those of the original attributes. The clusters let analysts select a subset of the attributes in the original dataset producing about the same classification accuracy as the full set of attributes while reducing the size of the dataset. Both techniques provide data analysts additional insight into a dataset.
Worcester Polytechnic Institute
Major Qualifying Project
All authors have granted to WPI a nonexclusive royalty-free license to distribute copies of the work, subject to other agreements. Copyright is held by the author or authors, with all rights reserved, unless otherwise noted.