0
$\begingroup$

I am about to start up a project on pattern recognition in a highdimensional dataset holding information on transactional salesdata for a company. In that manner I have decided to use the method of SOM to find clusters in the data.

However, I have wondered how it is possible to include the possibility of finding some kind of seasonal pattern? For instance, weekdays are, as far as I know, often modelled as binary. In this way it wont be possible to determine whether there if some certain customer behaviour in weekends, months, etc.

Do there exist a more sophisticated way of including dates, months and time?

Kind Regards

$\endgroup$
0
$\begingroup$

Assuming that the transactional data is timestamped to a resolution of (say) one second, you will need to bin it into to intervals of hours / days / weeks / months etc in order to form time-series vectors for each product / sales location / salesperson etc.

Seasonal behaviour is most often detected using auto-correlation and the use of a Correlogram. The use of an SOM will help you find clusters of similar sales patterns. Be careful how you normalise your data, though: whether you are more interested in the shape of the sales over time or the size of the sales over time will determine how you should normalise.

| cite | improve this answer | |
$\endgroup$
0
$\begingroup$

This was exactly what I was searching for https://ianlondon.github.io/blog/encoding-cyclical-features-24hour-time/

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ This is being automatically flagged as low quality, probably because it is so short. At present it is more of a comment than an answer by our standards. Can you expand on it? You can also turn it into a comment. $\endgroup$ – gung - Reinstate Monica Feb 6 '18 at 20:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.