Urgent : Linear Regression tool error : The field fit stats is not contained in the record

[AIPH123]

Hi All,

Am trying to run linear regression. When i run the flow, am getting an error as attached shown screen shot Saying " Fit Stats is not contained in the record", Can someone help me diagnose this.

Thank you.

Br,

Anil

[AIPH123]

Hi Roland,

Thank you for your input.

When i try to forecast and look at the forecast values, the values same for different periods as attached. Can you please let me know what causes this? I can see a same problem where people are facing in different threads, but i dont see any solution there. Can you please throw some light on this.

Thank you.

Br,

Anil

[RolandSchubert]

Hi Anil,

I could not analyze data in detail, but it seems, data provided for previous periods do not show a clear seasonality. I've added a TS Plot tool to the workflow, I think, it's helpful to analyze the results.

If you select some of the territory-segment-combinations, you'll usually find a pattern ike that (Browse tool attached to the "I" anchor):

  

The autocorrelation plot looks like this:

As you see, correlation between a period and a previous period decreases - even if it increases slightly, it shows no "seasonal pattern", which would look like this:

As there is no "seasonal pattern", changes are considered to be "remainder" (i.e. "error" or "random"). In addition, you'll so no long term trend, this leads to a more or less flat forecast.

Do you expect a seasonal pattern?

Best,

Roland

[AIPH123]

Hi Roland,

Yes, We have 3 levels of area splits i.,e Region, District and Territory. For example 1 region will have 5 districts and each district will have 7 territories and each territory will have 5 segments like Tier 1, 2,3,4 and 5 ., When we see the data on segments, i don't see a seasonality but when i look at the data on territories, i see a seasonality. Basically, summing all segments will give a territory data which is showing a seasonality.

But why will the segment data which has a variable trend in the past when forecast is constant? 

Br,

Anil

[RolandSchubert]

Hi Anil,

I see your problem ... unfortunately, the tool evaluates on territory-segment resulting in no seasonality and variance is considered to be random. The result is a flat forecast (no seasonality, no clear trend, all variance random => average forecast).

I would try to perform the process on territory (this should result in seasonality) and then allocate the result (total territory) to the segments by monthly shares last year or average share for last 3 years. What do you think?

Best,

Roland

[AIPH123]

Hi Roland,

Its seems, i don't have an option other than doing that on territory and then considering percentage of business on segments on historical data and splitting based on average of historical data. 

But problem is, if i average out at segments, that does not give a full picture because each year volumes from different segments are driven by various business drivers. I have to find some critical drivers and quantify them and do a weighted percentage allocations and  then split the forecast. 

Let me do it on total territory level forecast and see how my forecast look like.

Br,

Anil