🤖 XAUUSD AI Trading Bot

Industrial-grade MetaTrader 5 Expert Advisor powered by LightGBM and Smart Money Concepts (SMC)

Validated Performance: 66.2% win rate | 1.96 profit factor | 3,780% return (7-month backtest)

---

📊 Performance Summary

Metric	Result	Status
Win Rate	66.2%	✅ Excellent
Profit Factor	1.96	✅ Profitable
Max Drawdown	19.5%	✅ Under Control
LONG Accuracy	64.1%	⭐ Strong
SHORT Accuracy	73.0%	⭐⭐ Very Strong
Trades/Day	15.7	⚠️ Can optimize to 4-5
Test Period	7 months	Apr-Nov 2025

---

🎯 Features

AI-Powered Predictions

• LightGBM classifier with 68 features
• 55% confidence threshold for trade filtering
• ONNX format for MT5 integration
• Smart Money Concepts (SMC) quality scoring

Advanced Feature Engineering

• 21 Technical indicators (ATR, RSI, MACD, Bollinger Bands, etc.)
• 10 Market structure features (FVG, Order Blocks, Liquidity Sweeps)
• 8 Orderflow metrics (CVD approximation, delta divergence)
• 8 Volatility context features
• 6 Time-based features (session positioning)
• 1 SMC quality score (4-step validation)

Risk Management

• 5% equity risk per trade
• ATR-based dynamic stop-loss
• 25% max drawdown kill switch
• Trailing stop activation at +10%
• London-NY overlap trading only (13:00-16:59 UTC)

---

🚀 Quick Start

Prerequisites

• Python 3.11+
• MetaTrader 5
• 16GB RAM

Installation

Clone and setup: git clone https://github.com/andywarui/xauusd-ai-trading-bot.git cd xauusd-ai-trading-bot python -m venv .venv .venv\Scripts\activate pip install -r requirements.txt

text

Data Pipeline

python src/merge_yearly_data.py python src/validate_merged_data.py python src/filter_overlap.py python src/feature_engineering.py python src/create_labels.py

text

Model Training

python src/train_lightgbm.py python src/analyze_confidence.py python src/backtest_simple.py

text

ONNX Export (Python 3.11)

py -3.11 -m venv .venv_onnx .venv_onnx\Scripts\activate pip install lightgbm onnx onnxmltools python src/export_to_onnx_simple.py

text

---

📁 Project Structure

xauusd-ai-trading-bot/ ├── data/ │ ├── raw/ # Raw XAUUSD M1 data │ └── processed/ # Filtered and labeled ├── src/ │ ├── merge_yearly_data.py │ ├── filter_overlap.py │ ├── feature_engineering.py │ ├── train_lightgbm.py │ └── backtest_simple.py ├── python_training/models/ # Trained models ├── mt5_expert_advisor/Files/ # ONNX model (222 KB) └── docs/ # Documentation

text

---

🧪 Backtest Results

Configuration: $50 capital | 5% risk | 55% confidence | 149 days

Total Trades: 2,332 Win Rate: 66.2% Profit Factor: 1.96 Net Profit: $1,890 (3,780%) Max Drawdown: 19.5%

LONG: 1,787 trades | 64.1% WR | $1,262 profit SHORT: 545 trades | 73.0% WR | $627 profit

text

---

📈 Model Details

LightGBM Classifier

• Trees: 102
• Features: 68
• Training: 142,511 samples
• Testing: 35,628 samples
• Classes: SHORT | HOLD | LONG

Feature Categories: Technical (21) | Market Structure (10) | Orderflow (8) | Time (6) | Volatility (8) | Price Action (6) | SMC Score (1)

---

🎯 Roadmap

✅ Completed (62.5%)

• Data acquisition (178k bars)
• Feature engineering (61 features)
• Model training (LightGBM)
• ONNX export
• Backtest validation

⏳ In Progress

• MT5 EA development
• Risk management in MQL5
• Strategy Tester validation

🔮 Future

• Shadow testing (30 days)
• Live deployment
• Auto-retraining pipeline

---

🔧 Planned v2 Enhancements

• News filter (ForexFactory API)
• Time filters (avoid session opens/closes)
• Sentiment features (COT, DXY)
• Multi-timeframe confirmation
• Ensemble models

---

⚠️ Disclaimer

Educational purposes only. Trading carries substantial risk. Past performance ≠ future results. Test on demo accounts first.

---

📧 Contact

Author: Andy Warui
Repository: https://github.com/andywarui/xauusd-ai-trading-bot

---

📜 License

MIT License

---

⭐ Star this repo if you find it useful!