A New Imbalanced Learning and Dictions Tree Method for Breast Cancer Diagnosis

Abstract

In the most of standard learning algorithms it is presumed or at least expected that the distribution of data points in different classes of at-hand dataset are balanced; it means that there are almost the identical number of data points in each class. It is also implicitly presumed that the misclassification cost of each data point is a fixed value regardless of its class. The standard algorithms will fail to learn if the at-hand dataset is imbalanced. An imbalance dataset is the one that the distribution of data points among their classes is not identical; it means that the numbers of data points in different classes are considerably different. A well-known domain in that it is highly likely for each exemplary dataset to be imbalanced is patient detection. In such systems there are many clients while a few of them are patient and the all others are healthy. So it is very common and likely to face an imbalanced dataset in such a system that is to detect a patient from various clients. In a breast cancer detection that is a special case of the mentioned systems, it is tried to discriminate the patient clients from healthy clients. It should be noted that the imbalanced shape of a dataset can be either relative or non-relative. The imbalanced shape of a dataset is relative where the mean number of samples is high in the minority class, but it is very less rather than the number of samples in the majority class. The imbalanced shape of a dataset is non-relative where the mean number of samples is low in the minority class. This paper presents an algorithm which is well-suited for and applicable to the field of non-relative imbalanced datasets. It is efficient in terms of both of the speed and the efficacy of learning. The experimental results show that the performance of the proposed algorithm outperforms some of the best methods in the literature.