topic Re: Time Series Forecasting Per Account in Discussions

Time Series Forecasting Per Account

Dukowitz62 — Fri, 09 Jun 2023 00:37:57 GMT

We are trying to trouble shoot Time Series Forecasting in JMP. We have data that represents water consumption (WTR_CCF) per account (SP_ID) for each month & year from 2010-2021; the data table has over 59 million rows. We want to forecast water demand at the account level, but keep running into an error when we use the Time Series Forecast model that says “no unique time periods” We are using the M/Y column; however, because we have the same M/Y for each account, the model won't run. Can someone help troubleshoot this? We did a regression model as well using the other variables in the table, and have a prediction function saved to the data table…but how do we then use this to predict out 30 years and save the estimates as forecasted values in the time series table?

We have attached a tiny subset data table and images for the Predictive Model setup and resulting formula.

Re: Time Series Forecasting Per Account

peng_liu — Sat, 21 Aug 2021 02:16:54 GMT

What error message does the Time Series Forecasting platform say? Is it something like in this screenshot? If so, that means your series probably has duplicated timestamps, e.g. 07/2015 appears twice or more in Customer Y's series. Even a series or more has problem, the platform should be able to analyze other series that don't have problems.

Data quality is a challenge during forecasting many series. Group Summary of Time and Group Summary of data can help to reveal difficult-to-spot issues.

Re: Time Series Forecasting Per Account

Dukowitz62 — Fri, 03 Sep 2021 13:41:18 GMT

We were attempting to forecast for each customer which had data for each month. This is just not possible. So, yes, errors were definitely related to non-unique data for particular dates. We were able to successfully get a time series forecast by aggregating individual customers by customer category by month. This also solved the problem posed in my post from 08/31/2021:

Thank you.