Learning from imbalanced data: open challenges and future directions

Abstract

Despite more than two decades of continuous development learning from imbalanced data is still a focus of intense research. Starting as a problem of skewed distributions of binary tasks, this topic evolved way beyond this conception. With the expansion of machine learning and data mining, combined with the arrival of big data era, we have gained a deeper insight into the nature of imbalanced learning, while at the same time facing new emerging challenges. Data-level and algorithm-level methods are constantly being improved and hybrid approaches gain increasing popularity. Recent trends focus on analyzing not only the disproportion between classes, but also other difficulties embedded in the nature of data. New real-life problems motivate researchers to focus on computationally efficient, adaptive and real-time methods. This paper aims at discussing open issues and challenges that need to be addressed to further develop the field of imbalanced learning. Seven vital areas of research in this topic are identified, covering the full spectrum of learning from imbalanced data: classification, regression, clustering, data streams, big data analytics and applications, e.g., in social media and computer vision. This paper provides a discussion and suggestions concerning lines of future research for each of them.

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

“…In particular, we share the opinions expressed in the position paper (Krawczyk 2016). According to it while decomposing the multiple imbalanced classes, pairwise relations between two classes only may be a too strong over-simplification and they do not reflect more complex relations between several of classes, as one class influences several neighboring classes at the same time.…”

“…In this study, we decided to use the method where the type of example can be identified by analysing class labels of the k-nearest neighbours of this example. 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.…”

“…This formulation is justified by focus an interest in the most important class and its real-life semantics, like in medical diagnosis (distinguishing sick vs. healthy patients), spam detection (e.g., valid activity vs. malicious one), image recognition (target object vs. background), etc. Even if the dataset includes more majority classes, then they are aggregated into one global majority class as in most applications it is necessary to improve the recognition of the minority class -for yet stronger its justification see He and Ma (2013), Krawczyk (2016), Sun et al (2009), Weiss (2004. A straightforward discrimination between the minority class and the majority one have led to a development of several methods that take into account only this relation, such as random re-sampling.…”

“…The decomposition of the multi-class imbalanced dataset to a set of binary problems is a dominating strategy so far, see a review in (Krawczyk 2016). Researchers consider either one-vs-all classes or one-vs-one decompositions.…”

“…For their review see, e.g., (Branco et al 2016;He and Garcia 2009;He and Ma 2013;Krawczyk 2016). These methods are usually categorized in data level and algorithm level approaches.…”