Random Forest

A Random Forest is a supervised learning method for classification and regression of data, in which various decision trees are generated that are as different as possible. The values or classes resulting from the various decision trees (see also CART) are then combined to produce a result, which can often yield more accurate predictions than a single decision tree.

To construct a decision tree that classifies known data as effectively as possible, the attribute that best classifies the existing data is chosen at each branching (node). In the construction of decision trees in a Random Forest, only a random subset of the possible attributes is selected for each node, ensuring that different decision trees are created for classifying the data. Additionally, different datasets can be used, which are generated by leaving out or duplicating data from the existing dataset (Bagging).

The results from the different decision trees are then combined, e.g., as a weighted average, to form an overall result. By considering an ensemble of decision trees, this combined result contains insights from many individual decision trees, resulting in a better classification (see also Ensemble Methods). It is important that the various decision trees are as uncorrelated as possible, meaning they should not be too similar.

Random Forests are an efficient method for large datasets with many attributes, features, and training data. In the context of time series forecasting, the forecast is determined from the class identified by the Random Forest.