The Effectiveness of Using Diversity for Selecting Multiple Classifier Systems

A wealth of approaches exists to perform classification of items of interest. The goal of fusion techniques is to exploit complementary approaches and merge the information provided by these methods to provide a solution superior than any single method. Associated with choosing a fusion algorithm is the choice of algorithm or algorithms that will be fused. This decision is most often referred to as ensemble selection. Historically classifier ensemble accuracy has been used to accomplish this task. More recently research has focused on creating and evaluating diversity metrics to more effectively select ensemble members. This research focuses on the use of diversity as an ensemble selection methodology and explores the relationship between ensemble accuracy and diversity. Using a wide range of classification data sets, classification methodologies, and fusion techniques it extends current diversity research by expanding classifier domains before employing fusion methodologies; this is made possible with a unique classification score algorithm developed for this purpose. Correlation and linear regression techniques determine the relationship between examined diversity metrics and accuracy is tenuous and optimal ensemble selection should be based on ensemble accuracy