Identifying Mislabeled Instances in Classification Datasets
This addresses the costly and error-prone issue of manual data labeling for supervised machine learning, offering a practical solution for improving dataset quality across numerical, image, and natural language domains.
The paper tackles the problem of mislabeled instances in classification datasets by presenting a non-parametric end-to-end pipeline that identifies such errors with an average precision of over 0.84 when reviewing the top 1% recommendations on 29 datasets with added label noise.
A key requirement for supervised machine learning is labeled training data, which is created by annotating unlabeled data with the appropriate class. Because this process can in many cases not be done by machines, labeling needs to be performed by human domain experts. This process tends to be expensive both in time and money, and is prone to errors. Additionally, reviewing an entire labeled dataset manually is often prohibitively costly, so many real world datasets contain mislabeled instances. To address this issue, we present in this paper a non-parametric end-to-end pipeline to find mislabeled instances in numerical, image and natural language datasets. We evaluate our system quantitatively by adding a small number of label noise to 29 datasets, and show that we find mislabeled instances with an average precision of more than 0.84 when reviewing our system's top 1\% recommendation. We then apply our system to publicly available datasets and find mislabeled instances in CIFAR-100, Fashion-MNIST, and others. Finally, we publish the code and an applicable implementation of our approach.