Hi everyone, I dont' know if this is the correct subreddit for this question, but in case let me know where i should post.

I am facing a regression task where I have to predict the price of some diamonds using a tabular dataset. I know that highly correlated features need to be removed and I noticed that 4 features, namely "carat", "x", "y" and "z", have correlation above 0.9. The problem is the performance of the model. If I keep all variables, I get a mean squared error of 702188 , If I keep only one of the three, I get 3107246 . I am using a PoissonRegressor(alpha=0.001, max_iter=5000) because the target variable has mean equal to the standard deviation. I really don't understand why my model has this behaviour, do you have any idea? Thanks for any help