model problem: finding the 'good' observation in each set of observations
I'm somewhat new to rapidminer and need help / suggestions on how to setup a process for the following problem:
I have a pretty lage dataset (over 1 mio observations): there are 1-20 observations with the same key. Besides that they have 4 more numeric attributes.
Also there is one label (0 or 1) and it's 1 for just one observation in each 'key-group'.
In other words: for each key there is exactly one observation marked with a 1 and all other have a 0.
The challenge is to find that correlation between the other 4 attributes and the label.
If I just use a decision tree, then obviously it doesn't understand that all observations with the same key belong together.
Any advice would be really great - especially how I model that there is just one 'good' observation for each key.