How can I predict the best treatment to give to new patient?

Question

As part of a school project, I have to analyze a dataset with patients (with characteristics: sex, age, smoker 0/1, etc.) who received different treatments (one per patient) with a response to this treatment 1/0.

Ex :

Patient 1 | Man | 45 years | Smoker 0 | Diabetes 0 | Obesity 1 | Treatment A | Response 1
Patient 2 | Man | 20 years | Smoker 1 | Diabetes 0 | Obesity 0 | Treatment B | Response 0
Patient 3 | Man | 57 years old | Smoker 1 | Diabetes 1 | Obesity 0 | Treatment C | Response 0
Patient 4 | Women | 49 years old | Smoker 1 | Diabetes 0 | Obesity 0 | Treatment B | Response 0
Patient 5 | Women | 42 years old | Smoker 0 | Diabetes 0 | Obesity 1 | Treatment A | Response 0
Patient 6 | Women | 34 years old | Smoker 0 | Diabetes 0 | Obesity 0 | Treatment C | Response 1

I want to set up a model that will predict the best treatment (with the best probability of positive response) to put in place for each new patient who will arrive.

I thought about making a prediction model (starting with the random forest) of positive response per treatment which will give me a probability, and in the end take the treatment with the highest probability. Do you think I'm on the right track? Is there anything better to implement in this scenario?

I don't know how to test this model because the predicted treatment doesn't necessarily match the treatment that was actually given to the patient in my dataset, so I can't know the 0/1 response to this patient/treatment set at all shots.

Thanking you in advance if you have any ideas.

Answer 1

Your plan to use a random forest model to predict treatment based on patient characteristics is solid. Evaluate with cross-validation, do explore feature enhancements and tune hyperparameters. Also, check for class imbalance.

To validate your results try splitting your data into 60:20:20 - train, test and validation. Make sure your train is balanced. Good Luck!!