Fig. 1

Hierarchical meta-classification. An overview of our sampling-based ensemble approach for multiclass classification while addressing data imbalance. a Coarse classification: The top dashed rectangle corresponds to the meta-classification scheme used to separate records associated with the combined class consisting of stages 4 and 5 from those associated with stage 3. b Refinement step: The bottom dashed-rectangle corresponds to the classification step aiming to separate the combined class of advanced CKD stages (stages 4&5) into its two individual constituent classes (stage 4 vs. stage 5). In both steps, shaded rectangles represent the actual classifiers employed, the grey ovals denote the final classification outcome. The white oval in the Coarse Classification diagram represents the intermediate set of records, assigned to the combined class (stages 4&5) by the meta-classifier, which is further split in the refinement step. The input to the base-classifiers and to the refinement classifier consists of 455-dimensional vectors, representing the respective sets of patient records; the input to the meta-classifier comprises M dimensional vectors whose components correspond to the labels assigned by each of the M base-classifiers to the original (455-dimensional) vectors