The ultimate Fabric, Power BI, SQL, and AI community-led learning event. Save €200 with code FABCOMM.
Get registeredCompete to become Power BI Data Viz World Champion! First round ends August 18th. Get started.
I've just started to use the forecasting feature on some time series data in Power BI desktop and it all seems to work fine except that depending on the data the forecast is sometimes flat, and other times bumpy. I can't seem to figure out what is causing this, and I can't find any documentation or anything that explains it.
If I use my full dataset the forecast line is flat as seen here:
However, if I exclude some categories the line goes bumpy like this:
There doesn't seem to be any rhyme or reason (that I can figure out) as to why including or excluding some categories changes the forecast line to flat or bumpy. There are no blanks in the data or anything like that - each row in my dataset has a cumulative audience figure, each line has a date, etc.
Can anyone explain what is going on here? Thanks!!
Solved! Go to Solution.
I find the crux of this issue. When we use "Auto" Seasonality, it will detect as yearly. There is not too big difference on yearly changes, when this value change split into daily, it almost "flat" every day.
If you want to forcast it more accurate, you can change the Seasonality into a bigger value, like 12 (evaluate on monthly).
Regards,
Tough to say without sample data, formulas and relationships but it looks like the category you are excluding is some sort of null or blank? Perhaps that is messing with the calculations somehow.
I've attached the data I'm using but with some identifying details removed and I'm still getting the same issues with the forecasting line being flat. You'll notice that looking at individual categories only the two with the smallest number of data points actually results in a proper forecast line (H & I). If you combine multiple categories you get some combinations that will also produce a proper forecast line (e.g. A,C,D,F,G), but most combinations are just flat.
I'm not sure of the common way to share files on here but I've uploaded the .xlsx and .pbix to Jumpshare here:
http://jmp.sh/b/FUkBib3pXO2LjhMo8Cdz
Any help would be greatly appreciated as this forecasting is for an important project and I would at least like to understand why it's not working for the majority of the data, otherwise I look like a bit of an amatuer! =S
I find the crux of this issue. When we use "Auto" Seasonality, it will detect as yearly. There is not too big difference on yearly changes, when this value change split into daily, it almost "flat" every day.
If you want to forcast it more accurate, you can change the Seasonality into a bigger value, like 12 (evaluate on monthly).
Regards,
This is great! Thankyou so much. I should have checked that out, it kind of makes sense now in that the forecasts that were working properly were those that had greater change over the years.
Thanks again 🙂