Types of minority class examples and their influence on learning classifiers from imbalanced data

Abstract

Many real-world applications reveal difficulties in learning classifiers from imbalanced data. Although several methods for improving classifiers have been introduced, the identification of conditions for the efficient use of the particular method is still an open research problem. It is also worth to study the nature of imbalanced data, characteristics of the minority class distribution and their influence on classification performance. However, current studies on imbalanced data difficulty factors have been mainly done with artificial datasets and their conclusions are not easily applicable to the real-world problems, also because the methods for their identification are not sufficiently developed. In our paper, we capture difficulties of class distribution in real datasets by considering four types of minority class examples: safe, borderline, rare and outliers. First, we confirm their occurrence in real data by exploring multidimensional visualizations of selected datasets. Then, we introduce a method for an identification of these types of examples, which is based on analyzing a class distribution in a local neighbourhood of the considered example. Two ways of modeling this neighbourhood are presented: with k-nearest examples and with kernel functions. Experiments with artificial datasets show that these methods are able to re-discover simulated types of examples. Next contributions of this paper include carrying out a comprehensive experimental study with 26 real world imbalanced datasets, where (1) we identify new data characteristics basing on the analysis of types of minority examples; (2) we demonstrate that considering the results of this analysis allow to differentiate classification performance of popular classifiers and pre-processing methods and to evaluate their areas of competence. Finally, we highlight directions of exploiting the results of our analysis for developing new algorithms for learning classifiers and pre-processing methods.

Multi-class and feature selection extensions of Roughly Balanced Bagging for imbalanced data

“…For instance, if k = 5, the type of the example is assigned in the following way (Napierala and Stefanowski 2012;2016): 5:0 or 4:1 -an example is labeled as a safe example; 3:2 or 2:3 -a borderline example; 1:4 -labeled as a rare example; 0:5 -example is labeled as an outlier. This rule can be generalized for higher k values, however, results of recent experiments (Napierala and Stefanowski 2016) show that they lead to a similar categorization of considered datasets. Therefore, in the following study we stay with k = 5.…”

“…Recall that different difficulty factors could be considered: a fragmentation of the minority class into small disjuncts, overlapping of decision boundaries, presence of rare cases, outliers, noise (Stefanowski 2016a). Here we follow the methodology from Napieraha and Stefanowski (2012Stefanowski ( , 2016, where most of these data difficulty factors can be modeled by distinguishing the following types of examples: safe examples (located in the homogeneous regions populated by examples from one class only); borderline (placed close to the decision boundary between classes); rare examples (isolated groups of few examples located deeper inside the opposite class), or outliers.…”

“…For more details, the reader is referred to a recently published monograph (He and Ma 2013) covering the most representative issues and to the earlier systematic surveys, such as Chawla (2005), He andGarcia (2009), Sun et al (2009). A recent, comprehensive review of pre-processing methods could be found in Branco et al (2016) and their comparative studies are provided by Napierala and Stefanowski (2016), Van Hulse et al (2007).…”

“…Then, values start to fluctuate around certain levels. However, both datasets are the smallest ones as well as the distributions of the minority class are the most sparse and the most difficult ones (Napierala and Stefanowski 2016).…”

Exploring complex and big data

“…Also, even the more traditional learning tasks, such as training classifiers, are now reformulated in more demanding ways, for instance, by taking into account additional constraints or data properties, like unusual distributions of examples and/or imbalance of target classes (Fernández et al, 2017;Napierala and Stefanowski, 2016). Such enriched input to the induction process requires more advanced and complex algorithms.…”