The remaining minority classes are arranged as per their respective status of lowest minority class having I.R. > 1.5.
The minority classes are oftentimes the ones of interest (e.g., when they are associated with pathological conditions in patients), so methods for handling imbalanced datasets are critical.
It is interesting to notice that farthest-first beats d-Confidence on the majority classes (Tables 10) but, when all majority classes are found and only minority classes are left unexposed, d-Confidence reveals its ability to find rare instances.
The classes of this cluster (positive part) are then sorted in descending order of class size so that around 15%% of the datapoints of the minority classes are moved to the other cluster (negative part).
Ironically, minority classes are typically the most important to learn, because they may be associated with special cases.
In contrast with SVM-OAA, the ensemble version EnSVM-OAA helps to slightly improve the overall classification accuracy Acc, with possible sacrifice of two other evaluation metrics on most datasets, which means that classification accuracies between majority and minority classes are further increased.
To maximize performance, samples from the positive (minority) class were weighted to compensate for data imbalances, and model parameters were tuned in ten-fold cross-validation such that the minimum number of trees exceeded 1,000 [22], [23].
When the initial representation of the minority class is small, this hindrance is obviously more severe.
Furthermore, the class prior probabilities in this context are highly unequal (e.g. the distribution of the positive class is 0.1%% in the 1,000 classes classification problem), and over-sampling the minority class is very expensive.
Indeed, the resulting models were respectively trees with all the leaf nodes predicting the majority class and sets of CARs where the minority class was highly underrepresented, when not absent.
