How to optimize the parameters to make the RMSE fitting value close to 0 when predicting multiple output items through multiple input items?

[bodypumper]

The following is the practical process

1：The experimental data for the original concrete composition ratio is 1000 pieces, and the algorithm model used is the random forest regression model.

2：The following code is for creating models, training models, predicting target values, and optimizing Optuna.

The fitting index of RMSE is 27.781625571862275.

[nzw0301]

What about the training loss? If it's small, it will be overfitting. Otherwise, n_trials looks large enough to converges, so I suppose expanding the hyper-parameters of random forest or use different regressor (neural net?) might reduce the loss.

[nzw0301]

What about the training loss?

[bodypumper]

I am currently using Optuna for parameter tuning to avoid training losses. Is there any other way to solve my question?

[nzw0301]

I think it is not possible to say something without comparing training loss value and

RMSE value remains around 18.

[nzw0301]

To be clear, I meant sharing the training loss, not optimising training loss by optuna.

[bodypumper]

To be clear, I meant sharing the training loss, not optimising training loss by optuna.

First, I employed data preprocessing techniques, including standardization.
Subsequently, I utilized boosting methods, specifically a blend of decision trees and AdaBoostRegressor.
I am exploring alternative approaches to address the issue of training loss.