Artificial Neural Network as Ensemble Technique Fuser for Improving Classification Accuracy

Abstract

Ensemble learning is one of the highly accurate and robust learning approaches. In these approaches, different classifiers are used as an ensemble technique, and then the main questions that faces up is how to fuse the results of each individual classifier into a final decision. In this paper, we will propose a fuser based on an Artificial Neural Network to produce the final decision. our proposed approach has been experimentally validated on three public datasets (available in UCI repository). The first dataset is called Wisconsin Prognosis Breast Cancer (WPBC) dataset. This dataset has 35 attributes and 198 instances. The second dataset is called Breast Cancer Wisconsin (Original) DataSet. This dataset has 10 attributes and 699 instances. The third dataset is called Diabetic Retinopathy Debrecen DataSet. This dataset has 20 attributes with and 1151 instances. Our proposed approach gives higher accuracy in these datasets. The accuracy of our approach is 82.7% with the (WPBC) DataSet, 98.5% with the Breast Cancer Wisconsin (Original) DataSet and 75% with Diabetic Retinopathy Debrecen DataSet