I want to forecast some data. I have been playing and reading about ARIMA but I find myself not being able to implement it, so I changed direction. I am currently using the meta nodes that Knime has included in the time series section and I have been able to succesfully (with some error) implement a regression.
Now I want to further deepen into the forecasting. If the last day is “190” I want to be able to extend the forecast to, for example, 290. Something like this.
I was guessing I should use the Lag Column node but I am a bit lost right here.
I would also like to know more about time forecasting and I know there will be one course in october ([L4-TS] Introduction to Time Series Analysis - Online | KNIME) but I would like to know the do’s and dont’s, what method to use and what kind of data is best for what method, if anyone is willing to help.
Hi @jorgemartcaam , the Lag Column only shifts the rows of the column.
I’m not sure what your 3 columns represent. I’m guessing the first column is the Day of the Year? But for the other 2 columns, I have no idea what they are.
Are there any business rules or algorithms for the prediction?
The first column is the day of the year, like you said.
The second column represents the number of contacts made in a day. I.e.: there were 220 useful calls made the day 176 (2021-06-25).
The third column represents the prediction, in this case 264 calls.
For the prediction I am using the “Auto-Prediction” meta node that comes with the time series nodes in Knime. Regarding the use of algorithms and business rules, I am just using the regression with the lagged values of the contacts, days and hours. Nothing else.
Best regards!
