This article may be helpful.

Article Clustering scientific documents with topic modeling

If someone uses clustered data as inputs to classifiers, prediction accuracies of the classifiers must be high because he has already divited raw data to three clusters according to their similarity. The major problem is that labels provided by clustering may not match with true labels of raw data.

Mr Dong and Caitlin,

Thanks for your answers, yes you have a point. But, i found a solution in one paper. It totally depends up on the dataset, my data has one label and one id and many attributes. So, i need not worry about your problem of matching labels. First, i tried with different clustering algorithms and found one best clustering algorithm after getting accuracy using classification algorithms. So this method helped me finding the best clustering algorithms (one with high accuracy), when i have no idea which algorithm to choose for my dataset. I even used Geo coding to map and validate the cluster results and also Davies–Bouldin index to evaluate my analysis (number of clusters). This method is valid and helps in identifying best algorithm for a specific dataset.

Steps

1. Run clustering algorithms

2. Get accuracy using Classification algorithms (choose the best one)

3. Evaluate using Davies-Bouldin Index

4. Use Geo coding (in my case i have lat and long ) to validate the analysis

Regards,

Athiq