One-Sided Prototype Selection on Class Imbalanced Dissimilarity Matrices

Abstract

Abstract. In the dissimilarity representation paradigm, several prototype selection methods have been used to cope with the topic of how to select a small representation set for generating a low-dimensional dissimilarity space. In addition, these methods have also been used to reduce the size of the dissimilarity matrix. However, these approaches assume a relatively balanced class distribution, which is grossly violated in many real-life problems. Often, the ratios of prior probabilities between classes are extremely skewed. In this paper, we study the use of renowned prototype selection methods adapted to the case of learning from an imbalanced dissimilarity matrix. More specifically, we propose the use of these methods to under-sample the majority class in the dissimilarity space. The experimental results demonstrate that the one-sided selection strategy performs better than the classical prototype selection methods applied over all classes.

Denaturing high‐performance liquid chromatography and principal component analysis for identification of DNA point mutations in breast cancer and lymphoma samples

“…Random under sampling is a simple approach to reduce the number of samples in the majority class. This technique has been used for the classification of all kind of datasets …”

“…This technique has been used for the classification of all kind of datasets. [40][41][42][43] Figure 5 illustrates G mean performance versus whisker length. For subsets where minority class accounted for 11.1 to 55.6% of mutated samples (upper row), low G mean values were obtained when free parameter w in the classification criterion took values below 1; that is, as w decreased, all mutated samples were correctly classified, but the number FIGURE 5 G-mean performance versus whisker length for each PER3A assembled subset.…”

An Empirical Study of Oversampling and Undersampling for Instance Selection Methods on Imbalance Datasets

“…For the One-sided prototype selection method [18] we followed the steps proposed by the authors. First, we apply DROP3, CHC or IRB over the whole database.…”

“…The main problem is that the minority class always obtains lower accuracy than the majority class. Only a few works have been proposed to deal with the imbalance problem on instance selection [18], however, there are some techniques based on oversampling and undersampling that can be combined with IS methods to improve the accuracy of the minority class.…”

“…On the other hand, in [18] the One-sided method is proposed as an approach to instance selection over imbalanced datasets. The main idea of this work is to carry out instance selection only over the majority class leaving intact the minority class.…”