I am currently (after a break) getting back into KNIME and need your help. I couldn’t find anything like this in the Community Hub. I would like to perform the following analysis or prediction:
I have turnover data from various suppliers for the last 13 years (on years base), including possible bonus agreements. I would like to build a workflow that processes this data and provides me with future sales predictions for the next 2-3 years and, if possible, the corresponding bonus values.
I found sales forecasts in the Community Hub, but they don’t meet my requirements.
Can anyone help me with this? Perhaps someone has already built this workflow.
Hi rfreigel,
you are right!
Here are my test files with turnovers and bonus scales.
The expected output should be the “calculated” turnover for the next 3 years and the “expected” amount of achieved bonus in EUR for the next 3 years.
I hope you can help me.
All the best
René Bonus-Scales-Test-File.xlsx (12.4 KB) Test-File.xlsx (12.8 KB)
Here you go. Simple ARIMA model. I didn’t play with the ARIMA parameters. The workflow could probably be structured with a single branch with a loop but my approach was easier for a prototype. I don’t know anything about your business. It appears based on the data that turnover was affected by COVID. Its always good to start with some simple data exploration. I created a line chart with a linear trend line which raised my suspicion about COVID. I built the bonus structure with a Column Expressions node (below). Final thought - be careful about blindly accepting the predictions. Three year forecasts based only on historical outcomes are pretty iffy.
You are right. I already thought to add (at one point) affects from outside like costs for gas, fuel or electricity, inflation, tariffs, etc. but this is currently way out of my knowledge. I wouldn’t even know where to get that data right now. I’d probably ask ChatGPT.
One question regarding the prediction. I saw that the last both years show the same turnover and logically the same bonus. Do you know why both years got the same prediction?
@ReneStromberg you could check out these collections about Time Series. And I think you are right to think about external factors that will influence your numbers.
Covid will be difficult to handle. You might have to start a fresh series and see how this does perform.
@ReneStromberg As I said in my last post, I did not try to optimize the ARIMA parameters. Optimization is part art, part science. Domain knowledge is pretty important. Knime has a Parameter Optimization Loop which can be helpful. Nodepit has a component which purports to optimize ARIMA parameters.
I didn’t have much luck with either. I relied on my “domain knowledge” about the COVID downturn. Open the ARIMA learner nodes and change all three parameters to 1. I think that produces a reasonable outcome from your limited data. Google “ARIMA parameters” and study the function of the parameters. Same warning as before - be cautious about blindly trusting the output.
