A Lazy Ensemble Learning Method to Classification

Depending on how a learner reacts to the test instances, supervised learning divided into eager learning and lazy learning. Lazy learners endeavor to find local optimal solutions for each particular test instance. Many approaches for constructing lazy learning have been developed, one of the successful one is to incorporate lazy learning with ensemble classification. Almost all lazy learning schemes are suffering from reduction in classifier diversity. Diversity among the members of a team of classifiers is deemed to be a key issue in classifier combination. In this paper we proposed a Lazy Stacking approach to classification, named LS. To keep the diversity of classifiers at a desire level, LS utilizes different learning schemes to build the base classifiers of ensemble. To investigate LS’s performance, we compare LS against four rival algorithms on a large suite of 12 real-world benchmark datasets. Empirical results confirm that LS can statistically significantly outperform alternative methods in terms of classification accuracy.