- 期刊
Retinal Blood Vessel Segmentation Using Pareto Ensemble Prunning with Diversity
摘要
Ensemble learning way is effectively determined to form a group of base learners with the provided hypotheses combination, which improves performance of the traditional single classifier systems. How to improve the correct responses of base learners is one of the fundamental challenges in ensemble learning systems. Although Pareto Ensemble Pruning (PEP) can concurrently maximize the generalization performance and minimize the number of base learners in an ensemble system, it still has a question to be elucidated: diversity. Diversity is a necessary machine for high generalization capability in classifier ensemble. In this study, a novel Pareto Ensemble Pruning with Diversity (PEPD) approach is proposed on the basis of negative correlation learning and NSGA-II multi-objective genetic algorithm. There are three goals in PEPD: minimizing the error and size; maximizing the diversity. Experimental results on DRIVE data sets illustrate that the PEPD algorithm achieves the better performance than the other state-of-the-art approaches.