A novel intrinsic measure of data separability

In machine learning, the performance of a classifier depends on both model and separability/complexity datasets. To quantitatively measure separability datasets, in this study, we propose an intrinsic – Distance-based Separability Index (DSI), which is independent model. We then formally show that DSI can indicate whether distributions datasets are identical for any dimensionality. because consider situation different classes data mixed same distribution to be most difficult classifiers separate. And, verified effective by comparing it state-of-the-art measures using synthetic real (CIFAR-10/100). Having demonstrated DSI’s ability compare samples, our other studies used separability-based applications, such as measuring generative adversarial networks (GANs) evaluating results clustering methods.