Hello dear experts,
i'm new to KNIME and overall to Machine Learning field. So I have one question about KNIME and need one advice about choosing best algorithms for my tasks.
The data set I'm working with has following structure:
- Time, device_1(float)
- Time, device_2(float)
- Time, device_N(float)
And some other stuff like Time and Booleans for states (on, off). In the future it will have more than 100 different params.
=> The Target is to predict a sum of all float values for 1-5 minutes (each dataset = 1 second interval)
I tried to use 10000 to 100000 datasets for different algorithms like Linear and Polynomial Regression, Decision Tree, Random Forest etc. and always got different results compared to how high lag interval is. (Tested it with a lag interval from 10 to 30 -> means 10 to 30 seconds)
Question 1: Can you please give me an advice about best algorithm choice based on your option? (Every explanation will help me).
- After deploying a PMML to Predictive Plugin in another product, I have noticed that new incoming data don’t have any option to self-learn and actualize PMML statistics for future predictions on this new data basis.
Question 2: Is there a way to do some model with self-learning option in KNIME? If not, I would be thankful, if you tell me about another existing possibilities :)