An efficient randomised sphere cover classifier

摘要

This paper describes an efficient randomised sphere cover classifier (aRSC), that reduces the training data set size without loss of accuracy when compared to nearest neighbour classifiers. The motivation for developing this algorithm is the desire to have a non-deterministic, fast, instance-based classifier that performs well in isolation but is also ideal for use with ensembles. We use 24 benchmark datasets from UCI repository and six gene expression datasets for evaluation. The first set of experiments demonstrate the basic benefits of sphere covering. The second set of experiments demonstrate that when we set the a parameter through cross validation, the resulting aRSC algorithm outperforms several well known classifiers when compared using the Friedman rank sum test. Thirdly, we test the usefulness of aRSC when used with three feature filtering filters on six gene expression datasets. Finally, we highlight the benefits of pruning with a bias/variance decomposition.