Time series with multiple attributes

nearwick

Hello,
I'm a bit lost as to how exactly I should create a model for my data. I have a row for each day with multiple numeric attributes and the predicted attribute is also numeric. I want to predict the next value based on the previous data (let's say previous 5 days), but also the other attributes of the predicted row.
Example:
date         att1   att2   att3  val
1.1.2018    7       0.3    33   12
2.1.2018    0       0.6    31   14
3.1.2018   0.5      5       32   ?
I want to predict the value of "val" in the last row, knowing all the values in the previous two rows but also the values of all the other attributes in the last row.
I suspect windowing and other time series operators and techniques should be used, but I'm not exactly sure how. All sample processes for time series predict only one value, not using any other attributes. 

hughesfleming68

Hello nearwick,

You are correct that you will need to use the windowing operator and a predictor operator. Multivariate time series forecasts can be set up by declaring your val attribute as the label. Univariate and multivariate predictions are two separate problems. You can however use a multiply operator and do both forecasts in the same process.

SGolbert

Hi @nearwick

while multiple univariate predictions can do the job, a prediction based on multiple attributes could be better. I am talking for example on regressing att1 to att1, att2 and att3, i.e. vector autoregressive models. Those are usable in RapidMiner, albeit using a scripting operator.

Kind regards,

Sebastian