Skip to Main Content
Machine Learning algorithms are difficult to directly apply among data sets of high dimensionality. This paper examines application of hybrid algorithms to segment data models to enable a higher level of accuracy. Our process begins with the reduction of our input parameter sets through the derivation of dominant characteristics. Using these characteristics, ranges are determined in which to segment our model. Each set is then used to train a predictive model using Machine Learning techniques. One major attribute of our application framework is to support an interchangeable set of algorithms for each stage. This process is demonstrated by estimating stated incomes from an automotive financing application for purpose of predictive modeling. We conclude that by applying our segmented hybrid framework we can achieve substantial improvements in accuracy over pure Machine Learning applications.
Date of Conference: March 30 2009-April 2 2009