1
$\begingroup$

I have a Customer data (20 periods) which contains Sales of each customer by Period P1,P2,P3 ... P20) (each period =14 days)

Is there anyway to predict P21 period ? Can I used ARIMA or ETS keeping in mind that the data has more zero values then non zeros for each customer. ?

Sample Data for one Customer

Improve this question
$\endgroup$

2 Answers 2

Reset to default
1
$\begingroup$

When a time series has "lots" of zeroes , one can approach it with tools that are referred to as Intermittent Demand Models also know as Sparse Data Models essentially dealing with two random variables ...the rate between demands and the interval between demands . With only 3 non-zeroes little can be done BUT if you have a more populated series there is hope.

Google "croston's method intermittent demand forecasting" to learn more

I and others have previously commented on these models https://stats.stackexchange.com/search?q=user%3A3382+croston

Improve this answer
$\endgroup$
2
  • $\begingroup$ I have a data set for only 20 periods for each customer, is there any snippet of code to implement on my data set in R or Python ? $\endgroup$
    – Usman Rafiq
    Commented Oct 12, 2019 at 9:12
  • $\begingroup$ can't precisely help you there. $\endgroup$
    – IrishStat
    Commented Oct 12, 2019 at 11:22
0
$\begingroup$

Going for an ARIMA/ETS for this is like taking bazooka for nothing.

Basically it could be a binary problem translation and you have a probability of 3/20 to have a prediction different from 0.

But on this kind of series, times series techniques won't help you ...

If it was a multivariate problem, it would be more interesting though.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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