I have a data set with three different features (x1, x2, x3) and I am going to use a regression model to predict y based on the features.
x3 is the total amount of money that a customer invest and y is the interest for that customer. So, for all cases y < x3. x1 and x2 are other information about the customer.
Rather than predicting y, I tried to predict y/x3 and I got much better results.
In my new model, I have tried to predict y/x3 = f(x1, x2, x3) using XGB model. The reason that I get better results is because of the reduction in the skewness in the data.
Do you think that this leads to data leakage? Please note that I did a lot of feature extraction based on x1, x2, x3, and use them for both models.