# Classification followed by clustering

Suppose i have a labelled training set with 1's and 0's already labelled for a binary classification problem. Suppose i wanted to use a clustering algorithm to classify. So i form clusters and then see if 1's and 0's tend to cluster well together. How would i do it?

Could i remove the labels create clusters and then see the proportions of 1's to 0's in each cluster? Then i could use this as a classifier. So if a cluster has a larger proportions of 1's any new item that needs to be classified would be also assigned a 1, otherwise a 0. Is this approach commonly used? What are the potential drawbacks?