I'm using xgboost to build a regression tree. I"m doing a GridSearch to find the optimal parameters. My training set has around 800 instances and I'm doing a CV = 3. Once the model is built, I see some issues in the residual plots. This is how the residual plots look like.
So There is an obvious decreasing pattern in the residual graph. Considering the fact that xgboost is very robust model, what can be wrong here?