Long story shortCreated a SSAS mining structure forecasting inventory usage/volume. Set it up, built and ETL to load it to SSAS and then pull it down with the SSAS transformer in SSIS using DMX. Overcame challenges here and there. The moment of truth! It was set up, even had built a nice dashboard etc.......Finally we started to review the data it made sense, the forecast was by quarter to help wash out some seasonality. Then an ah ha moment. Let's compare to other primative forms of forecasting, exponential smoothing and even just using the average of the last two periods. Keep in mind I ran it using ARIMA, ARTxp and Mixed, on the BI version. The findings were EXTREMELY disappointing. The average of the last two period killed ARIMA, ARTXP and Mixed. I dropped my data into minitab comparing the sets in a box plot, the outliers that were present from ARIMA, ARTxp or Mixed was way outside the maximum quartile. 3-7 times outside the whiskers. BRUTAL!I am going back now to use the average of the last 2 quarters for the first forecast period and then let ARIMA take over. Seriously disappointing results.......
↧