How to perform Normalization on Call Details Record to perform k-Mean Clustering I'm new to data mining and currently doing mining project on telecom customer segmentation (based on profile and call details record). I have gender, age, call time and call duration and have to perform k-mean clustering. My question is:- How to normalize data to perform clustering. Any reference or suggestion are welcomed.
 A: K-means is inherently distance based, so any normalization implicitly includes assumptions about the relationship between the different variables. Also I notice that your dataset only have 3 continuous variables (and one binary), so you can actually visualize the data relatively easily. Any clusters that are so robust to be meaningful to your client should be visible to your highly evolved pattern recognition system (your brain). I imagine that your dataset is large, so you might have to subsample or estimate a density function from the original data.
You also have the problem that one of your variables is circular (clock time), so you can't use that one directly in a k-mean algorithm because it's metric distance based, so you should convert it into a space in which the 'distance' between 11:59 pm and 12:01 am is the same as between 12:01 and 12;03. I think it might work to map into rectangular coordinates (but that adds a dimension to your data).
For more information visit this link http://www.youtube.com/watch?v=22bqfmiZAKU
