1
$\begingroup$

I need to perform clustering and classification of time series of weekly sales of different products. My data are weekly sales of different products in differest areas (stores). The challenges on this problem are:
- Time-series are usually short: 10-52points(weeks).
- Time-series may have a lot of zeros - sparce data. Products do not sell every week.
- Not all products start to sell on the same date. This can result in time-shifted time-series. Even the same typical lifecycle of a product can be time-shifted in calendar along different stores.
- Sales may have noise such as random events, promotions etc.

A sample of data is like this:

20140105,prod1,store1,5
20140119,prod1,store1,10
20140126,prod1,store1,2
....
20140105,prod1,store2,2
20140112,prod1,store2,3
....
20140112,prod2,store3,4
20140126,prod2,store3,7

Can somebody share any insight on how to do this? Is it good to use a method such as DTW to compare time-series?If so, how am I going to handle the timeshifts? As for the implementation R seems a good way to go. Which packages would you recommend?

$\endgroup$