A Multiple Resampling Method for Learning from Imbalanced Data Sets

Re-Sampling methods are commonly used for dealing with the class-imbalance problem. Their advantage over other methods is that they are external and thus, easily transportable. Although such approaches can be very simple to implement, tuning them most effectively is not an easy task. In particular, it is unclear whether oversampling is more effective than undersampling and which oversampling or undersampling rate should be used. This paper presents an experimental study of these questions and concludes that combining different expressions of the re-sampling approach is an effective solution to the tuning problem. The proposed combination scheme is evaluated on imbalanced subsets of the Reuters-21578 text collection and is shown to be quite effective for these problems. Introduction In a conceptlearning problem, the data set is said to present a class imbalance if it contains many more examples of one class than the other. Such a situation poses challenges for typical classifiers such as Decision Tree Induction Systems or Multi-Layer Perceptrons that are designed to optimize overall accuracy without taking into account the relative distribution of each class (Japkowicz & Stephen 2002; Estabrooks 2000). As a result, these classifiers tend to ignore small classes while concentrating on classifying the large ones accurately. Unfortunately, this problem is quite pervasive as many domains are cursed with a class imbalance. This is the case, for example, with text classification tasks whose training sets typically contain much fewer documents of interest to the reader than on irrelevant topics. Other domains suffering from class imbalances include target detection, fault detection, or fraud detection problems, which, again, typically contain much fewer instances of the event of interest than of irrelevant events. ∗ Corresponding author