Which method combines multiple decision trees sequentially to improve predictive accuracy?

The method that combines multiple decision trees sequentially to improve predictive accuracy is known as boosted trees. Boosting is an ensemble technique that involves training a series of weak learners (often decision trees) where each new tree is trained to correct the errors made by the previous trees. This sequential learning process helps to enhance the model's predictive performance by focusing more on the observations that were misclassified or poorly predicted in earlier iterations.

Boosted trees work by adjusting the weights of the training instances based on their errors, allowing subsequent trees to concentrate more on these difficult cases. As a result, the final model is a weighted sum of all the weak learners, leading to better accuracy than that of individual decision trees.

In contrast, bagging (or bootstrap aggregating) builds multiple independent models and averages their predictions, which is different from the sequential approach of boosting. Random forests also use multiple decision trees but do so in parallel rather than in a sequential manner, focusing on reducing variance rather than improving accuracy through error correction like boosting does. Support vector machines, on the other hand, are a different class of models based on finding optimal hyperplanes for classification tasks and do not involve decision trees at all.