1
$\begingroup$

I have generated a model for predicting the future input trend with a sample data using linear regression in Knime. I want to validate the model using a different data set.

Suppose the data set I used for creating the model came from the sensor of device A. The prediction is accurate for the dataset from device A. Lets say I saved the model as a PMML file called A. Can I use the same PMML file A to predict values of Device B (Values of B are not comparable). If not, should I create a different model for all data set that I have?

Question

How do I combine all the models generated so that it predicts any given data set? Is it possible?

$\endgroup$
  • $\begingroup$ I think you answered yourself by "values of B are not comparable". Learning for predictions is based on a fundamental assumption which is the data for prediction has the same joint distribution as the data for learning. This is the link between those processes. $\endgroup$ – rapaio Mar 9 '16 at 10:27
  • $\begingroup$ OKAY. In that case how do I combine all the models into a single model? I am using time series fr prediction and there are chances that the data from the same device are not always same. $\endgroup$ – Vini Mar 9 '16 at 13:17
1
$\begingroup$

I think you answered yourself by "values of B are not comparable". Learning for predictions is based on a fundamental assumption which is the data for prediction has the same joint distribution as the data for learning. This is the link between those processes.

Now, if you want to handle that in a meaningful way you have to know somehow the source type. In your example the device type. One way would be to introduce the device type as a different column in your data set, so that the model can have the chance to differentiate between source type. Obviously you have to have training data for all the device types. Supposing you have 2 device types, A and B. Your training data should have some columns for the signal and also a factor column with type A and B. Also you have to have enough instances of data with type A and type B.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.