Fast growing and interpretable oblique trees via logistic regression models

摘要: The classification tree is an attractive method for as the predictions it makes are more transparent than most other classifiers. widely accepted approaches to tree-growth use axis-parallel splits partition continuous attributes. Since interpretability of a diminishes grows larger, researchers have sought ways growing trees with oblique they better able observations. focus this thesis grow in fast and deterministic manner propose making them interpretable. Finding good computationally difficult task. Various authors proposed doing by either performing stochastic searches or solving problems that effectively produce at each stage tree-growth. A new approach finding such restricts attention small but comprehensive set splits. Empirical evidence shows found cases. When observations come from number classes, empirical can be grown matter seconds. As main strength trees, important logistic regression, well-founded variable selection techniques introduced trees. This allows concise so interpretable directly grown. In addition this, cost-complexity pruning ideas which were developed been adapted make major practical component providing oblique.tree package R casual users experiment way was not possible before.