Ensemble Training with Classifiers Selection Mechanism

The target of classifiers ranking and selection in the ensemble learning method is to find a small size of learners to improve the performance while saving the computational resources. It is challenging to select classifiers from an ensemble since ensemble learning produces indistinguishable classifiers. We present a novel technique for ranking and selecting classifiers in an ensemble of classifiers using the accuracy margin and diversity measurement (AMDM) method to resolve this issue. Accuracy margin, computed from classifiers’ output, plays a crucial role in ranking and selecting classifiers while other conventional measures fail to way out. Experimental results on three different benchmarks (biomedical datasets such as colon cancer, leukemia, and lymphoma) show that AMDM performs better or equal to that of other single and ensemble classifier based algorithms.