Predictive Modeling in Power BI?

[helloWorld]

I want to incorporate statistical analyses into my Power BI dashboards for forecasting purposes.

My goal is to create a forecast that recalculates in response to filters clicked on and off by the user.  This would allow for some really powerful "What if" scenario visualization.

To achieve this, would the calculations absolutely need to be performed within Power BI? 

Or is it possible for Power BI to send a filtered data set to another software (like a cloud based analytics software), which could:

         a) Calculate the model results

         b) Write the new calculations to a database

If that setup is possible, then I could set up a query in Power BI to grab the model results in real time.

My question is:  is it possible to do predictive modeling from within Power BI (besides using really complex DAX patterns)?  Also, is it possible to do predictive modeling using other software in a way that the calculations would respond to the users' filter selections in Power BI?

[Greg_Deckler · Accepted solution]

@helloWorld - I am not aware of being able to feed anything from Power BI to another program in the way you are describing. I did create a blog article on forecasting with Power BI:

http://community.powerbi.com/t5/Community-Blog/Correlation-Seasonality-and-Forecasting-with-Power-BI...

[helloWorld]

Thanks for your help!

[ImkeF]

For datasets up to 150k rows you can use some R-trickery in Power BI: http://www.thebiccountant.com/2015/12/28/how-to-export-data-from-power-bi-and-power-query/

On SQL-server-level there is a full R-integration - actually seems to be the fastest R-engine on the market at the moment.

@Greg_Deckler 's DAX-forecasts provide some good ground.

M can do all that and a bit more, as it also natively handles recursions and is more versatile in general (but mostly not as fast as DAX). FInd a pattern for a TREND-function here: https://github.com/ImkeF/pquery/blob/master/Stat.Trend.m

As of today we have parameters in PBI giving the users the ability to define filters: So if we can find some volunteers to put the code for all statistical functions together ... 🙂