Hi all, This is my first time asking about prediction nodes.
In you opinion, how can I predict the grand total for November and December?
In my KNIME DB there are every single record that create this pivot, so a have a huge amount of data, but this is all new form me and I don’t know where to start.
@gcas How many years of data do you have? If you have enough data you could perform a times series analysis which could account for seasonality and trend. If your data is limited, you can perform some type of regression although I’d be very careful trusting the results.
might depend on some factors but, some kind of moving averages, or linear interpolation or sth. like this
Then there are arima models when dealing with time forecasting. There are some workflows on the hub for this as well.
Here’s a simple moving aggregation workflow. You can play with the widow size. Frankly, I wouldn’t use it. Just because one can develop a workflow that runs doesn’t mean it produces usable output. What do your numbers represent and are they subject to probable seasonality? You said that you have 2 years’ data. You might be able to develop a time series workflow that’s better than moving aggregation but its unlikely. Good data with ample sample size is as important as choice of analytic tools. How to Predict the Next 2 Months.knwf (20.0 KB)
If you’re willing to change your historical data which are apparent outliers, the new branch I added to the workflow may produce better predictions. This is more a judgment call than a modelling issue. Your call. How to Predict the Next 2 Months 2.knwf (196.5 KB)