I would like to understand which is the best Machine Learning approach (regression, classification, ...) in the following scenario: I have a dataset with hundreds of people, each of them with a credit amount. The goal is to predict the recovered amount for each person, in order to sort according to this output value (I have the recovered amount in the training set). The goal is to identify the people that will pay the highest amount, not the highest ratio of debt.
I think that three different approaches could be used:
- train a regression algorithm, directly on the recovered amount for the training set. In such a way, the prediction output will be directly the amount. With this method, my fear is that there will be predicted values with an higher value than the maximum possible one
- train a regression algorithm on the ratio of recovered over the total, and then multiply such quantity for the real amount. In my opinion, this method is more robust
- train a classification algorithm, and then use the output score (no labels) to multiply the amount. It is somehow similar to the previous one.
Do you have any idea or something in the literature that could help me? Thanks!