Ensemble-based hybrid approach for breast cancer data

Abstract

Classification of datasets with characteristics such as high dimensionality and class imbalance is a major challenge in the field of data mining. Hence to restructure data, a synthetic minority over sampling technique (SMOTE) was chosen to balance the dataset. To solve the problem of high dimensionality feature extraction, principal component analysis (PCA) was adopted. Usually a single classifier is biased. To reduce the variance and bias of a single classifier an ensemble approach, i.e. the learning of multiple classifiers was tested. In this study, the experimental results of a hybrid approach, i.e. PCA with SMOTE and an ensemble approach of the best classifiers obtained from PCA with SMOTE was analyzed by choosing five diverse classifiers of breast cancer datasets.