Cascade generalization: Is SVMs´ inductive bias useful?

The problem of choosing the best classification algorithm for a specific problem domain has been extensively researched. This issue was also the main motivation behind the ever increasing interest in ensemble methods since 1992. In this paper, we propose a new method for classifiers´ fusion, which integrates cascade generalization and voting techniques. The proposed method utilizes two learning algorithms only, with an SVM as base level classifier, while a different classification algorithm is utilized at the meta level. This is then followed by a final voting stage. Our results show that the proposed method, even though simple, is a promising classifier ensemble, which compares favorably to other well established ensemble methods.