customer segmentation using a large data with many zeros
I am looking for clustering techniques that are tolerant to large
datasets (500,000 unique customers with transaction records).
I basically would like to conduct customer segmentation based on their
transaction history - what they bought, how often they visited stores,
demographics etc. And transaction part of the data is binary: 1 if
they bought, let's say, fruits etc.
Now the problem is that
1. transaction part includes lots of zeros
2. not every variables are continuous
Polychoric correlations might be useful for the second part, but I am
not sure how to go about the first one.
I appreciate if anyone could give me advice.