Ensembles of classifier models typically deliver superior performance and canoutperform single classifier models given a dataset and classification task athand. However, the gain in performance comes together with the lack incomprehensibility, posing a challenge to understand how each model affects theclassification outputs and where the errors come from. We propose a tightvisual integration of the data and the model space for exploring and combiningclassifier models. We introduce a workflow that builds upon the visualintegration and enables the effective exploration of classification outputs andmodels. We then present a use case in which we start with an ensembleautomatically selected by a standard ensemble selection algorithm, and show howwe can manipulate models and alternative combinations.
展开▼
