Ensembles of random sphere cover classifiers

• We developed novel ensemble algorithms for generalised instance based classifiers.
• The base classifier used is a randomised sphere cover classifier.
• The ensemble take advantage of the random selection of spheres and the parameters to add diversity to the ensemble.
• The subspace method for ensemble design is used successfully for the randomised sphere cover.
• We tested the algorithms on various data sets including a case study using gene expression data sets.

We propose and evaluate a new set of ensemble methods for the randomised sphere cover (RSC) classifier. RSC is a classifier using the sphere cover method that bases classification on distance to spheres rather than distance to instances. The randomised nature of RSC makes it ideal for use in ensembles. We propose two ensemble methods tailored to the RSC classifier: αβRSE, an ensemble based on instance resampling and αRSSE, a subspace ensemble. We compare αβRSE and αRSSE to tree based ensembles on a set of UCI data sets and demonstrate that RSC ensembles perform significantly better than some of these ensembles, and not significantly worse than the others. We demonstrate via a case study on six gene expression data sets that αRSSE can outperform other subspace ensemble methods on high dimensional data when used in conjunction with an attribute filter. Finally, we perform a set of bias/variance decomposition experiments to analyse the source of improvement in comparison to a base classifier.