Use ensemble techniques like bagging to train multiple models on different subsets of the dataset, where each subset is created by selecting different combinations of numerical and categorical features.

Use ensemble techniques like boosting to train models sequentially, where each model focuses on the instances with different combinations of numerical and categorical features and tries to predict their values more accurately.

Use ensemble techniques like stacking to combine the predictions of multiple models, each trained on a different preprocessing method for handling numerical and categorical features, such as encoding or normalization.

All of the above