The fuzzy support vector data description based on tightness for noisy label detection

Abstract Machine learning is a data-driven approach and as machine learning research advances, noisy labels have become a widely studied issue. Noisy labels can significantly reduce the accuracy of supervised classification models, making it important to address this problem. In this study, we propose a new method for detecting noisy labels in datasets. This method first utilizes a deep pre-trained network to obtain a feature set from the image data. We then introduce a membership degree based on tightness into the support vector data description model to detect noisy data in the dataset. To further verify the accuracy of our method, we construct two noise sets: the density peak noise set and the random noise set. The experimental findings indicate that our proposed algorithm outperforms traditional support vector data description algorithms in detecting noisy labels. Moreover, we use confidence to correct noisy labels in the data. After correcting for noisy labels, the accuracy of image classification is significantly improved.