Building CatBoost Trading Models from Market Features in MQL5
Summary
This tutorial explains how to frame a trading problem for CatBoost gradient boosting without requiring Python or R expertise. It distinguishes manually written trading rules from supervised learning: predictors encode market information, while a target labels the desired action or outcome. Suggested predictors include price and indicator values, volatility, time and event features, levels, and data from related instruments or timeframes. For an example, the article proposes generating binary labels from a simple moving-average signal and preparing predictor and target samples for training.
It describes CatBoost as an ensemble of decision trees and discusses exporting data, training, validation, and examining predictor importance. The author recommends generating samples in an Expert Advisor to better reproduce live data timing, catch logic errors, and reduce accidental use of future information. The article presents a practical introduction rather than a rigorous machine-learning treatment. Predictor importance depends on both the model and the sample, and the excerpt gives no evidence that the example model is profitable out of sample; careful validation and target design remain essential.
Key ideas
- Supervised trading models use predictors as inputs and a target variable to represent the desired action or outcome.
- Candidate predictors can describe price, indicators, volatility, time, events, levels, and related markets.
- A simple existing strategy can generate binary labels for an initial classification task.
- Generating samples in an Expert Advisor can help reproduce live data timing and identify look-ahead errors.
- Predictor-importance measures depend on the sample and should not be treated as definitive evidence of general usefulness.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.