Learning with local models

摘要: Next to prediction accuracy, the interpretability of models is one fundamental criteria for machine learning algorithms. While high accuracy learners have intensively been explored, still poses a difficult problem, largely because it can hardly be formalized in general way. To circumvent this often find model hypothesis space that user regards as understandable or minimize user-defined measure complexity, such obtained describes essential part data. interesting parts data, unsupervised has defined task detecting local patterns and subgroup discovery. In paper, problem classification formalized. A multi-classifier algorithm presented finds global essentially used with almost any kind base learner provides an interpretable combined model.