Iranian cancer patient detection using a new method for learning at imbalanced datasets

Abstract

Most of standard learning algorithms presume or at least expect that distributions governed on the different classes of dataset are balanced. Also they presume that the misclassification cost of each data point is equal without considering its class. These algorithms fail to learn at the imbalanced datasets. Cancer detection is a well-known domain in which it is very common to face imbalanced class distributions. This paper presents an algorithm which is suit to this field, in both speed and efficacy. The experimental results show that the performance of the proposed algorithm outperforms some of the best methods in the field. © 2011 Springer-Verlag.