Have you considered using some machine learning to produce some symbolic regression with a form of GP?

I was running random forest on some datasets. So i need to keep only imp variables and run the algo. I dont have much experience in this.