Sales forcast


  • Hi,

    I would like to do a sales forcast for a restaurant involing factors like past and future weather data and take into account the impact of events in the neighborhood and holidays.

    How can I accomplish that in ssas datamining?

    I started with a the time series algo....and tried to find the relationship between sales and weather through the decision tree algo. My problem is how to combine these algorithm? Or can I accomplish all of the requirements with one of them?

    It would be awesome if you could help me out!


    Thursday, July 12, 2012 9:40 AM


All replies

  • You should be able to build your model with Microsoft Time Series Algorithm. 

    You can have your data in the form similar to

    Date                           Sales                   Chance of Rain                       Special Event Attendance

    01/01/2001                5000                       40                                                 500

    01/02/2001                3000                       30                                                   0

    01/03/2001                2500                       30                                                   0

    02/14/2001                6000                       0                                                    100

    When you create a model, there is an algorithm parameter that specifies what algorithm to use. You need to use ARTXP algorithm, not ARIMA (ARTXP does cross-predictions, this way predictions of one series values are affected by other series).

    Tatyana Yakushev []

    Thursday, July 12, 2012 4:56 PM
  • Hi Tatyana thanks for your help!

    The issue I have with this is, that I can not influence upcoming events with this method. For instance some events a random and are not on the exact same day each year! There is no way I can tell that the algorithm right? Same with the weather: I got weather data measurement for the last couple of years and would like to use  the actual weather forcast in my sales predicion! If I do a cross prediction it implicitly forcasts the weather automatically right?

    I wanna do a forcasts like that:

    For instance: Next week in general: sunny and warm each day , tuesday has a special event in one of my stores + it will rains. I think the time series algo can not help with that right? because it has no input option for future events?


    Friday, July 13, 2012 8:36 AM
  • Microsoft time series algorithm allows you to provide values for one series to get more accurate predictions for another series.

    (there is even sample viewer where you can do this kind of analysis without writing DMX )

    Here is how it is done (look at Example 3: Adding New Data and Using EXTEND_MODEL_CASES) and

    Tatyana Yakushev []

    Saturday, July 14, 2012 12:37 AM
  • Hi Tatyana,

    I cannot make this work by the examples you have provided. As I understand, EXTENT_MODEL_CASES allows you to attach data to the of the training set rather than predicting it. 

    Could you please give me a "quick" example how a simple model and DMX query would look like?

    Thanks in advance!

    Tuesday, July 23, 2013 2:07 PM