Train 7 models, each on a binary datasets separated by "has feature" and "does not have feature". The "has feature" dataset has the feature you are hoping the model will pickup on and the other is filled with the rest of the dataset. Some how find a way to combine these 7 models into one model or make a system to cast the predictions in a useful way. Maybe use the confidence scores or something.
Train 7 models, each on a binary datasets separated by "has feature" and "does not have feature". The "has feature" dataset has the feature you are hoping the model will pickup on and the other is filled with the rest of the dataset. Some how find a way to combine these 7 models into one model or make a system to cast the predictions in a useful way. Maybe use the confidence scores or something.