Dear community,
I m having issues with the ARIMA-Learner-Node.
I would like to create a predictive forecast for sales of hardware, for the next 12 months. I tried using ARIMA-nodes, but I get strange results. Especially concerning the AR-Order (see Screenshot 1). The model predicts exactly same amout of sales as it is in the training-data - ALWAYS for the amount of months written here in the AR-Order. Is this how it is supposed to be? Like this it is hard to interpret R^2 with numeric scorer…
Can you help me, is there a better way for ARIMA-Sales-Forecasting?
This ARIMA extension is in the labs and is not production ready yet. We encourage our users to use verified components from the time series category (in your case ARIMA Learner and ARIMA Predictor) or the SARIMA/X Learner and Predictor nodes from the new timeseries extension. Currently it is also in labs but is mature enough for use in this case.
I would encourage you to use SARIMA for improved forecasting since the model incorporates for the seasonal component as well in the target column.
Did I get it right - you suggest NOT to use ARIMA Learner-NODE and ARIMA-Predictor-NODE, but to use ARIMA Learner-COMPONENT and ARIMA Predictor-COMPONENT instead?
or even better to use SARIMA nodes