Hands-On Ensemble Learning with Python

Improving random forest

In an attempt to improve our model, we try to restrict its overfitting capabilities, imposing a maximum depth of 3 for each tree. This results in considerable performance improvement as the model achieves an MSE of 17.42 and a Sharpe value of 0.17. Further restricting the maximum depth to 2 improves the MSE score slightly more to 17.13, but reduces its Sharpe value to 0.16. Finally, increasing the ensemble's size to 50, using n_estimators=50, produces a considerably better model, with an MSE of 16.88 and a Sharpe value of 0.23. As we have only used the original feature set (20 lags of return percentages), we wish to also experiment with the expanded dataset we utilized in the boosting section. By adding the 15-day ...