Using Candlestick Patterns as Features for CatBoost Market Predictions

Summary

This article describes a workflow for turning single-candle patterns into machine-learning inputs. It defines patterns using rules on open, high, low and close prices, including thresholds for doji bodies, shadow lengths and body size. Those rules feed an indicator and a collection script that stores detected patterns for model training; the described deployment uses a CatBoost classifier in an MQL5 trading robot.

The document outlines a pipeline from pattern detection and data collection to model training and EA predictions, with supporting tools for data handling and ONNX model use. It discusses the need for instrument-specific thresholds and the ambiguity involved in distinguishing visually similar formations. Although the stated aim is to assess whether patterns improve AI-based trading, the supplied excerpt gives no measured performance or evidence that the approach beats markets. Candlestick definitions encode heuristics, and predictive value would need careful out-of-sample evaluation.

Key ideas

• Candlestick patterns can be represented as rule-based features derived from OHLC values.
• Thresholds for body size and shadow proportions make pattern detection explicit but may need instrument-specific adjustment.
• The described workflow collects detected patterns, trains a CatBoost classifier and deploys predictions through an MQL5 trading robot.
• Similar candle shapes can fit multiple pattern definitions, making consistent rules important.
• The excerpt provides no performance results demonstrating that the features improve trading outcomes.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.