Meta-classification of one-class classification models using ranking correlation and nearest neighbor

Abstract

Meta-learning is a technique to learn the learning process. In which, classification of machine learning (ML) models are underexplored direction. One important question is how to generalize various ML models as an input to a meta-classifier. To address this question, this paper advocates treating all ML models as one-class classification(OCC) models, and proposes a meta-classification algorithm of OCC models. The proposal represents OCC models as normality rankings and classifies them using nearest-neighbor and ranking-correlation metrics. The experiment classifies OCC models, where classes correspond to training datasets, algorithms, and hyperparameters. Meta-classification showed high accuracy for distinguishing 1) OCC models trained from dissimilar datasets and 2) four OCC algorithms trained from similar datasets. In addition, the discussion highlights that the classification of OCC models is essentially the classification of datasets that treats multiple samples as a single input. This function is exploratory analyzed using sleeping records. The proposed method can provide a unified solution for classifying OCC models, datasets, and rankings. Source code is uploaded to the public repository https://github.com/ToshiHayashi/ClassOCC.