none
Not getting Accurate forecasting in Time Series Using SSAS RRS feed

  • Question

  • I have been trying to do time series forecasting using SSAS on a monthly sales data and my forecast values for first two  months come good but from third month on words it varies. I have set my Parameters as below:

    Forecast_method=ARIMA

    Periodicity_hint={12,2}

    Prediction_Smoothing=1

    Auto_Detect_Periodicity=0.8

    Kindly Help me in this.

    Tuesday, December 6, 2016 12:04 PM

All replies

  • Prediction_Smoothing is only relevant when using the Mixed Algorithm.  I would consider omitting the Prediction_Smoothing.  

    I noticed you have your periodicity hint set to 12,2 (annualizing the cycle and a 2 month cycle)?  Is this what you intended?  I would consider omitting the Periodicity Hint altogether unless you really do have true seasonal cycles.

    Auto_Detect_Periodicity normally defaults at .6 I would consider washing it out altogether and try 0.0.  

    I found that with ARIMA predicts better from SSAS with less constraints.  

    I hate to say it but depending on your data, sometimes just averaging the last two periods gives you a more accurate forecast.  

    HTH

    Thursday, December 22, 2016 10:39 PM