B0581
Title: A comparison of classifiers for multiclass classification models with imbalanced datasets
Authors: Silvia Golia - University of Brescia (Italy) [presenting]
Maurizio Carpita - University of Brescia (Italy)
Abstract: Three categorical classifiers are considered which can be used with a multi-class target variable, that is a variable that admits k non-overlapping classes and the units are to be classified into one, and only one, of them. By categorical classifier, the procedure is implied which, starting from the probabilities assigned to all the categories by a suitable method, probabilistic classifier, transforms these probabilities into a single class. The three classifiers are the Bayes Classifier (BC), which assigns, based on the probabilistic classifier, a unit to the most likely class, and two alternatives, that is the max difference classifier (MDC) and max ratio classifier (MRC), which both involve the observed frequencies. Previous work demonstrated that in terms of Macro Recall and F-score measures and stability in the face of increasing class imbalance, MDC and MRC are better alternatives to BC. Currently, the role of the sample size is investigated, the choice of the probabilistic classifier and the presence of balanced/imbalanced dichotomous explanatory variables in the performances of the three categorical classifiers and to verify if the superiority of MDC and MRC over BC continues to hold.
