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

[RolandSchubert · Accepted solution]

Hi @AIPH123 ,

could you provide a bit more information, e.g. number of predictor variables? Is any output provided at output anchor "O" (i.e. the model)?
Is there is a linear relation between predictor variable(s) and target variable? 

Best,

Roland

[AIPH123]

Hi Roland,

Attached flow for your reference. As this is the first time am using linear regression, can you please look into the data and help me out with the approach please. Am trying to find the linear trend for volume A in the attachment. I cracked Fit_Stat error by just selecting the numeric data type in predictor variables. Also, i have one more challenge here, How do i calculate linear trend on each specific territory here? i have 230+ territories in the attachment, when i apply linear regression directly, it is calculating on all the territories, how can i avoid this as i need to find the trend on each individual territory. Please let me know.

Thank you.

Br,

Anil

[RolandSchubert]

Hi Aril,

I wonder if linear regression is the right approach. I my understanding, you use historic values for volume to predict future values, there are no predictor variables ("drivers" for volume). Did you consider using Time Series tools such as ARIMA or ETS? They calculate trends based on previous values of the same measure. 

Do you want to calculate trend for territory (segments aggregated) or within the territories for each segment separately? 

Best,

Roland 

[AIPH123]

Hi Roland.

I dint explore Arima or ETS. I will explore them now. Thank you for the input. My thought was to separate segments and calculate trend for each territory example like calculating the trend for Tier 1 separately and there by other segments separately  because that reduces the complexity.

Please advise if this approach make sense.

Br,

Anil.

[RolandSchubert]

Maybe the attached workflow is helpful to get an idea, how to proceed using ARIMA or ETS and TS Model Factory to calculate a large number of territories.

[AIPH123]

Hi Roland,

Can you please upload with 2019.4 version. Am not able to open this.

Br,

Anil

[RolandSchubert]

Hi Anil,

I've changed the workflow to 2019.4. You'll have to install TS Forecast Factory and TS Model Factory, you can also download from Gallery TS-Model-Factory  and TS-Forecast-Factory.

Best,

Roland

[AIPH123]

Hi Roland,

Thank you for your time. This is really helpful.

will the forecast factory automatically consider seasonality? 

Also, when i see output, i can see that my forecast is same for different sub periods i.,e i have specified to foecast for 6 months, it gives me same value 

can you also please explain

1. how do i interpret confidence intervals?

2.please explain "Use only covariance (Arima only) option in Arima tool

Thank you.

Br,

Anil

Br,

Anil

[RolandSchubert]

Hi Anil,

I'll try to answer the questions.

Seasonality is automatically considered by the Forecast Factory tool, if it's identified within the Model Factory tool, so if the result is a linear forecast, it's calculated based on the model created by the TS Forecast Factory. Unfortunately, it's not possible to customize model parameters (you could do this in the ARIMA tool) to fine tune seasonality. 

The confidence interval of 95% means, you can expect there is a only a 5% chance that the real value will not be in the range between upper and lower values around the forecast calculated by the model. 

Basically, ARIMA is an autoregressive moving average model and uses only the series itself to predict future values, In addition, you have the option to include he linear effect that an additional series has on the series (would make sense, if you want to predict sales for one product and there is an additional measure you are able to control - e.g. future prices or number of stores the product is sold).

Hope, this explains it. Let me know, if you need more detail.

Best,

Roland