FS_criterion
A measure to evaluate multi-labeled clustering systems
Article's Abstract:
This paper proposes an approach to evaluate the performance of multi-labeled clustering systems. These systems assign multiple labels to each object in a dataset based on their characteristics. Explicitly speaking, each sample can be assigned to some true class-labels by the expert, and the same sample can be assigned to some cluster-labels by the clustering system at the same time. However, evaluating the quality of these systems is challenging. One of the challenges in evaluating multi-labeled clustering systems is that traditional measures, such as class and cluster Entropy, do not consider multi-labeling. To address this, the paper proposes a measure called the FS criterion. This criterion can be used to evaluate not only the performance of the whole clustering system but also the performance of each cluster or class.