Training and testing data in machine learning

Question

I want to train data using K-means algorithm and then test it over another similar kind of data removing only one column. I am new to machine learning, don't understand where does the prediction part take place? We are just giving data and testing the accuracy. How can we apply the algorithm on test data (which obviously will be different) to predict the value of the missing attribute?

Answer 1

Unsupervised learning is used with the K-means clustering technique. Because you are not attempting to predict something, unsupervised learning does not employ labels. Instead, you're looking for a mechanism to arrange your data into clusters based on similar traits.
In Supervised Learning, the purpose of test (and frequently validation) sets is to verify the generalization properties of your model in order to avoid over-fitting. However, since you don't know the real clusters of the data in unsupervised learning, you can't evaluate this. As a result, employing a test set is pointless.