Random Forests for Forex: Bagging, Voting, and Regression Trees
Summary
The article explains random forests as ensembles of decision trees trained on resampled data and randomized feature subsets. For classification, it combines tree outputs by majority vote; for regression, it averages predictions. It also describes extending a decision tree classifier into a regressor by using the mean target value at leaves and variance reduction to choose splits.
The author compares a single tree with a ten-tree forest on the same dataset and reports improved training and test accuracy for the forest, then discusses applying the approach in a forex Expert Advisor. The article also demonstrates regression-tree fitting on an airfoil-noise dataset. These examples illustrate implementation rather than establish durable trading performance: the excerpt provides no detailed forex performance figures, and notes risks including overfitting, class imbalance, sensitivity to parameters, limited interpretability, and longer training time. Training a larger forest delayed trading while the trees were fitted.
Key ideas
- Random forests combine many decision trees trained on different resampled observations.
- Feature randomness and bootstrap sampling diversify trees, while voting or averaging combines their predictions.
- A regression tree predicts the mean target in each leaf and can select splits using variance reduction.
- The article reports better accuracy for a forest than a single tree on its demonstration dataset.
- Random forests can still overfit and may be costly to train or difficult to interpret.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.