Pattern Aided Classification
Guozhu Dong, Vahid Taslimitehrani · 2016
This paper makes several contributions to research on classification. First, it introduces a new style of classifiers, namely pattern aided classifiers (PXC), each defined by several pattern and group-specific-classifier pairs. A PXC uses patterns as conditions and it applies a group-specific classifier only to data instances satisfying its associated pattern. Second, it introduces a new classification algorithm, called Contrast Pattern Aided Classification (CPXC), for learning accurate PXCs. Experiments over multiple benchmark datasets confirm that CPXC often builds significantly more accurate classifiers than traditional classification algorithms. Third, it introduces the technique of opportunity-guided boosting and the concept of conditional classifier ensembles, and it provides insight on why certain datasets are very challenging to traditional classification algorithms.