I’d like to ask about a case when we would like to predict the best class of some input variable, so that the probability of event will be maximized.
For example the advertisement type for a given customer, which will maximize the probability of purchase. In collected data we have many various ad types, customer descriptors and informations if the purchase was made or not.
One solution which comes to my mind is to treat the ad type as an input variable, train regular probability model, then make predictions with all ad type configuration and then pick one giving the best estimation.
What are the other options?