What does a higher level of predictive accuracy in trees usually require?

A higher level of predictive accuracy in decision trees is generally linked to a larger dataset. This relationship exists because more data provides a richer set of examples for the model to learn from, allowing it to generalize better and make more accurate predictions. With a larger dataset, the decision tree has access to a variety of cases and scenarios, which helps in capturing the underlying patterns in the data more effectively.

While higher complexity of the model can sometimes lead to improved performance, it also runs the risk of overfitting to the noise in the dataset, especially if the dataset is small. Similarly, including more features may not always enhance the model's accuracy; it might complicate the model without yielding substantial improvements. Lastly, while having more trees in a forest (in the case of random forests) can certainly help in achieving better performance, the baseline improvement in predictive accuracy starts with the quantity and quality of the data used for training the tree model. Thus, a larger dataset is fundamental for attaining higher levels of predictive accuracy in trees.