Selecting locally specialised classifiers for one-class classification ensembles

摘要: One-class classification belongs to the one of novel and very promising topics in contemporary machine learning. In recent years ensemble approaches have gained significant attention due increasing robustness unknown outliers reducing complexity learning process. our previous works, we proposed a highly efficient one-class classifier ensemble, based on input data clustering training weighted classifiers clustered subsets. However, main drawback this approach lied difficult time consuming selection number competence areas which indirectly affects members ensemble. paper, investigate ten different methodologies for an automatic determination optimal They roots model clustering, but can be also effectively applied task. order select most useful technique, their performance multi-class problems. Numerous experimental results, backed-up with statistical testing, allows us propose fully method tuning clustering-based ensembles.