Learning from Imbalanced Data

In this chapter, we consider the classification of imbalanced data. When a dataset presents an imbalance between its classes, that is, an uneven distribution of observations among them, the classification task is inherently more challenging. Traditional classification algorithms (see Sect. 2.2 ) tend to favour majority over minority class elements due to their incorrect implicit assumption of an equal class representation during learning. As a consequence, the recognition of minority instances is hampered. Since minority classes are usually the ones of interest, custom techniques are required to deal with such data skewness. We study them in this chapter.