OCFSP: self-supervised one-class classification approach using feature-slide prediction subtask for feature data

One-class classification (OCC) is a machine learning problem where training data has only one class. Recently, self-supervised OCC algorithms have been increasing attention. These train the model for pretext tasks and use error OCC. However, these are specialized images, applying them to feature not practical or appropriate such purpose. The motivation of this study apply data. For purpose, paper proposes an approach using feature-slide prediction (FSP) subtask (OCFSP). main originality FSP subtask, which first In particular, proposed method creates self-labeled dataset by generating additional vectors with slide original self-annotating as number slides. Such applied multi-class classifier predict Since learns from class, accuracy seen class higher relative unseen classes. Accordingly, could be made FSP. methods experimented imbalanced-learn, covtype, kddcup datasets. OCFSP shows fair few given. addition, relatively fast testing speed, unlike image Therefore, bottleneck considered memory size, difference between Source code uploaded at https://github.com/ToshiHayashi/OCFSP