|
|
---
|
|
|
language: en
|
|
|
license: mit
|
|
|
library_name: sklearn
|
|
|
tags:
|
|
|
- trading
|
|
|
- finance
|
|
|
- gold
|
|
|
- xauusd
|
|
|
- forex
|
|
|
- algorithmic-trading
|
|
|
- smart-money-concepts
|
|
|
- smc
|
|
|
- xgboost
|
|
|
- machine-learning
|
|
|
- backtesting
|
|
|
- technical-analysis
|
|
|
- multi-timeframe
|
|
|
- intraday-trading
|
|
|
- high-frequency-trading
|
|
|
datasets:
|
|
|
- yahoo-finance-gc-f
|
|
|
metrics:
|
|
|
- accuracy
|
|
|
- precision
|
|
|
- recall
|
|
|
- f1
|
|
|
model-index:
|
|
|
- name: xauusd-trading-ai-smc-daily
|
|
|
results:
|
|
|
- task:
|
|
|
type: binary-classification
|
|
|
name: Daily Price Direction Prediction
|
|
|
dataset:
|
|
|
type: yahoo-finance-gc-f
|
|
|
name: Gold Futures (GC=F)
|
|
|
metrics:
|
|
|
- type: accuracy
|
|
|
value: 80.3
|
|
|
name: Accuracy
|
|
|
- type: precision
|
|
|
value: 71
|
|
|
name: Precision (Class 1)
|
|
|
- type: recall
|
|
|
value: 81
|
|
|
name: Recall (Class 1)
|
|
|
- type: f1
|
|
|
value: 76
|
|
|
name: F1-Score
|
|
|
- name: xauusd-trading-ai-smc-15m
|
|
|
results:
|
|
|
- task:
|
|
|
type: binary-classification
|
|
|
name: 15-Minute Price Direction Prediction
|
|
|
dataset:
|
|
|
type: yahoo-finance-gc-f
|
|
|
name: Gold Futures (GC=F)
|
|
|
metrics:
|
|
|
- type: accuracy
|
|
|
value: 77.0
|
|
|
name: Accuracy
|
|
|
- type: precision
|
|
|
value: 76
|
|
|
name: Precision (Class 1)
|
|
|
- type: recall
|
|
|
value: 77
|
|
|
name: Recall (Class 1)
|
|
|
- type: f1
|
|
|
value: 76
|
|
|
name: F1-Score
|
|
|
---
|
|
|
---
|
|
|
|
|
|
# XAUUSD Multi-Timeframe Trading AI Model
|
|
|
|
|
|
## Files Included
|
|
|
|
|
|
### Core Models
|
|
|
- `trading_model.pkl` - Original daily timeframe XGBoost model (85.4% win rate)
|
|
|
- `trading_model_15m.pkl` - 15-minute intraday model (77% validation accuracy)
|
|
|
- `trading_model_1m.pkl` - 1-minute intraday model (partially trained)
|
|
|
- `trading_model_30m.pkl` - 30-minute intraday model (ready for training)
|
|
|
|
|
|
### Documentation
|
|
|
- `README.md` - This comprehensive model card
|
|
|
- `XAUUSD_Trading_AI_Paper.md` - **Research paper with academic structure, literature review, and methodology**
|
|
|
- `XAUUSD_Trading_AI_Paper.docx` - **Word document version (professional format)**
|
|
|
- `XAUUSD_Trading_AI_Paper.html` - **HTML web version (styled and readable)**
|
|
|
- `XAUUSD_Trading_AI_Paper.tex` - **LaTeX source (for academic publishing)**
|
|
|
- `XAUUSD_Trading_AI_Technical_Whitepaper.md` - **Technical whitepaper with mathematical formulations and implementation details**
|
|
|
- `XAUUSD_Trading_AI_Technical_Whitepaper.docx` - **Word document version (professional format)**
|
|
|
- `XAUUSD_Trading_AI_Technical_Whitepaper.html` - **HTML web version (styled and readable)**
|
|
|
- `XAUUSD_Trading_AI_Technical_Whitepaper.tex` - **LaTeX source (for academic publishing)**
|
|
|
|
|
|
### Performance & Analysis
|
|
|
- `backtest_report.csv` - Daily model yearly backtesting performance results
|
|
|
- `backtest_multi_timeframe_results.csv` - Intraday model backtesting results
|
|
|
- `feature_importance_15m.csv` - 15-minute model feature importance analysis
|
|
|
|
|
|
### Scripts & Tools
|
|
|
- `train_multi_timeframe.py` - Multi-timeframe model training script
|
|
|
- `backtest_multi_timeframe.py` - Intraday model backtesting framework
|
|
|
- `multi_timeframe_summary.py` - Comprehensive performance analysis tool
|
|
|
- `fetch_data.py` - Enhanced data acquisition for multiple timeframes
|
|
|
|
|
|
### Dataset Files
|
|
|
- **Daily Data**: `daily_data.csv`, `processed_daily_data.csv`, `smc_features_dataset.csv`, `X_features.csv`, `y_target.csv`
|
|
|
- **Intraday Data**: `1m_data.csv` (5,204 samples), `15m_data.csv` (3,814 samples), `30m_data.csv` (1,910 samples)
|
|
|
|
|
|
## Recent Enhancements (v2.0)
|
|
|
|
|
|
### Visual Documentation
|
|
|
- **Dataset Flow Diagram**: Complete data processing pipeline from raw Yahoo Finance data to model training
|
|
|
- **Model Architecture Diagram**: XGBoost ensemble structure with decision flow visualization
|
|
|
- **Buy/Sell Workflow Diagram**: End-to-end trading execution process with risk management
|
|
|
|
|
|
### Advanced Formulas & Techniques
|
|
|
- **Position Sizing Formula**: Risk-adjusted position calculation with Kelly Criterion adaptation
|
|
|
- **Risk Metrics**: Sharpe Ratio, Sortino Ratio, Calmar Ratio, and Maximum Drawdown calculations
|
|
|
- **SMC Techniques**: Advanced Order Block detection with volume profile analysis
|
|
|
- **Dynamic Thresholds**: Market volatility-based prediction threshold adjustment
|
|
|
- **Ensemble Signals**: Multi-source signal confirmation (ML + Technical + SMC)
|
|
|
|
|
|
### Performance Analytics
|
|
|
- **Monthly Performance Heatmap**: Visual representation of returns across all test years
|
|
|
- **Risk-Return Scatter Plot**: Performance comparison across different risk levels
|
|
|
- **Market Regime Analysis**: Performance breakdown by trending vs sideways markets
|
|
|
|
|
|
### Documentation Updates
|
|
|
- **Enhanced Technical Whitepaper**: Added comprehensive visual diagrams and mathematical formulations
|
|
|
- **Enhanced Research Paper**: Added Mermaid diagrams, advanced algorithms, and detailed performance analysis
|
|
|
- **Professional Exports**: Both documents now available in HTML, Word, and LaTeX formats
|
|
|
|
|
|
## Multi-Timeframe Trading System (Latest Addition)
|
|
|
|
|
|
### Overview
|
|
|
The system has been extended to support intraday trading across multiple timeframes, enabling higher-frequency trading strategies while maintaining the proven SMC + technical indicator approach.
|
|
|
|
|
|
### Supported Timeframes
|
|
|
- **1-minute (1m)**: Ultra-short-term scalping opportunities
|
|
|
- **15-minute (15m)**: Short-term swing trading
|
|
|
- **30-minute (30m)**: Medium-term position trading
|
|
|
- **Daily (1d)**: Original baseline model (85.4% win rate)
|
|
|
|
|
|
### Data Acquisition
|
|
|
- **Source**: Yahoo Finance API with enhanced intraday data fetching
|
|
|
- **Limitations**: Historical intraday data restricted (recent periods only)
|
|
|
- **Current Datasets**:
|
|
|
- 1m: 5,204 samples (7 days of recent data)
|
|
|
- 15m: 3,814 samples (60 days of recent data)
|
|
|
- 30m: 1,910 samples (60 days of recent data)
|
|
|
|
|
|
### Model Architecture
|
|
|
- **Base Algorithm**: XGBoost Classifier (same as daily model)
|
|
|
- **Features**: 23 features (technical indicators + SMC elements)
|
|
|
- **Training**: Grid search hyperparameter optimization
|
|
|
- **Validation**: 80/20 train/test split with stratification
|
|
|
|
|
|
### Training Results
|
|
|
- **15m Model**: Successfully trained with 77% validation accuracy
|
|
|
- **Feature Importance**: Technical indicators dominant (SMA_50, EMA_12, BB_lower)
|
|
|
- **Training Status**: 1m model partially trained, 30m model interrupted (available for completion)
|
|
|
|
|
|
### Backtesting Performance
|
|
|
- **Framework**: Backtrader with realistic commission modeling
|
|
|
- **Risk Management**: Fixed stake sizing ($1,000 per trade)
|
|
|
- **15m Results**: -0.83% return with 1 trade (conservative strategy)
|
|
|
- **Analysis**: Models show conservative behavior to avoid overtrading
|
|
|
|
|
|
### Key Insights
|
|
|
- ✅ Successfully scaled daily model architecture to intraday timeframes
|
|
|
- ✅ Technical indicators remain most important across all timeframes
|
|
|
- ✅ Conservative prediction thresholds prevent excessive trading
|
|
|
- ⚠️ Limited historical data affects backtesting statistical significance
|
|
|
- ⚠️ Yahoo Finance API constraints limit comprehensive validation
|
|
|
|
|
|
### Files Added
|
|
|
- `train_multi_timeframe.py` - Multi-timeframe model training script
|
|
|
- `backtest_multi_timeframe.py` - Intraday model backtesting framework
|
|
|
- `multi_timeframe_summary.py` - Comprehensive performance analysis
|
|
|
- `trading_model_15m.pkl` - Trained 15-minute model
|
|
|
- `feature_importance_15m.csv` - Feature importance analysis
|
|
|
- `backtest_multi_timeframe_results.csv` - Backtesting performance data
|
|
|
|
|
|
### Next Steps
|
|
|
1. Complete 30m model training
|
|
|
2. Implement walk-forward optimization
|
|
|
3. Add extended historical data sources
|
|
|
4. Deploy best performing intraday model
|
|
|
5. Compare intraday vs daily performance
|
|
|
|
|
|
## Model Description
|
|
|
|
|
|
This is an AI-powered trading model for XAUUSD (Gold vs US Dollar) futures, trained using Smart Money Concepts (SMC) strategy elements. The model uses machine learning to predict 5-day ahead price movements and generate trading signals with high win rates.
|
|
|
|
|
|
### Key Features
|
|
|
- **Asset**: XAUUSD (Gold Futures)
|
|
|
- **Strategy**: Smart Money Concepts (SMC) with technical indicators
|
|
|
- **Prediction Horizon**: 5-day ahead price direction
|
|
|
- **Model Type**: XGBoost Classifier
|
|
|
|
|
|
## Romeo (V5) — Ensemble model
|
|
|
|
|
|
Romeo (codename V5) is the latest ensemble model combining tree-based learners (XGBoost / LightGBM) and an optional Keras head. The artifacts live in `models_romeo/` and include a canonical feature list used by the backtester to align unseen data.
|
|
|
|
|
|
Artifacts
|
|
|
- `models_romeo/trading_model_romeo_daily.pkl` — ensemble artifact (joblib) with `models`, `weights`, and `features` keys.
|
|
|
- `models_romeo/romeo_keras_daily.keras` — optional Keras model file when included in training.
|
|
|
- `models_romeo/MODEL_CARD.md` — this model's card with evaluation and transparency notes.
|
|
|
|
|
|
Evaluation (selected run on unseen daily data)
|
|
|
- Initial capital: 100
|
|
|
- Final capital: 484.8199
|
|
|
- CAGR: 0.0444
|
|
|
- Annual volatility: 0.4118
|
|
|
- Sharpe: 0.3119
|
|
|
- Max Drawdown: -47.66%
|
|
|
- Total trades: 3610
|
|
|
- Win rate: 49.47%
|
|
|
|
|
|
Uploading to Hugging Face
|
|
|
-------------------------
|
|
|
There is a helper script to upload the model artifacts to Hugging Face Hub:
|
|
|
|
|
|
1. Install dependencies:
|
|
|
```bash
|
|
|
pip install huggingface_hub
|
|
|
```
|
|
|
|
|
|
2. Set your HF token in the environment (Windows cmd.exe):
|
|
|
```cmd
|
|
|
set HF_TOKEN=hf_YourTokenHere
|
|
|
```
|
|
|
|
|
|
3. Upload:
|
|
|
```cmd
|
|
|
python v5\upload_model_v5_to_hf.py --repo-name your-username/romeo-v5 --model-dir models_romeo
|
|
|
```
|
|
|
|
|
|
The script will create the repo (if it doesn't exist) and upload all files from `models_romeo/`.
|
|
|
|
|
|
Usage example
|
|
|
-------------
|
|
|
Load the artifact and run predictions:
|
|
|
|
|
|
```python
|
|
|
import joblib
|
|
|
artifact = joblib.load('models_romeo/trading_model_romeo_daily.pkl')
|
|
|
features = artifact['features']
|
|
|
# prepare X matching features
|
|
|
# model usage depends on artifact['models'] layout; check MODEL_CARD.md for details
|
|
|
```
|
|
|
|
|
|
Notes & Next Steps
|
|
|
------------------
|
|
|
- Position sizing is simplified in the backtester; consider implementing fixed-risk sizing before live use.
|
|
|
- Consider re-running the robustness scan using the M2M metric as primary evaluation (recommended).
|
|
|
|
|
|
- **Accuracy**: 80.3% on test data
|
|
|
- **Win Rate**: 85.4% in backtesting
|
|
|
|
|
|
## Intended Use
|
|
|
|
|
|
This model is designed for:
|
|
|
- Educational purposes in algorithmic trading
|
|
|
- Research on SMC strategies
|
|
|
- Backtesting trading strategies
|
|
|
- Understanding ML applications in financial markets
|
|
|
|
|
|
**⚠️ Warning**: This is not financial advice. Trading involves risk of loss. Use at your own discretion.
|
|
|
|
|
|
## Training Data
|
|
|
|
|
|
- **Source**: Yahoo Finance (GC=F - Gold Futures)
|
|
|
- **Period**: 2000-2020 (excluding recent months for efficiency)
|
|
|
- **Features**: 23 features including:
|
|
|
- Price data (Open, High, Low, Close, Volume)
|
|
|
- Technical indicators (SMA, EMA, RSI, MACD, Bollinger Bands)
|
|
|
- SMC features (Fair Value Gaps, Order Blocks, Recovery patterns)
|
|
|
- Lag features (Close prices from previous days)
|
|
|
- **Target**: Binary classification (1 if price rises in 5 days, 0 otherwise)
|
|
|
- **Dataset Size**: 8,816 samples
|
|
|
- **Class Distribution**: 54% down, 46% up (balanced with scale_pos_weight)
|
|
|
|
|
|
## Performance Metrics
|
|
|
|
|
|
### Model Performance
|
|
|
- **Accuracy**: 80.3%
|
|
|
- **Precision (Class 1)**: 71%
|
|
|
- **Recall (Class 1)**: 81%
|
|
|
- **F1-Score**: 76%
|
|
|
|
|
|
### Backtesting Results (2015-2020)
|
|
|
- **Overall Win Rate**: 85.4%
|
|
|
- **Total Return**: 18.2%
|
|
|
- **Sharpe Ratio**: 1.41
|
|
|
- **Yearly Win Rates**:
|
|
|
- 2015: 62.5%
|
|
|
- 2016: 100.0%
|
|
|
- 2017: 100.0%
|
|
|
- 2018: 72.7%
|
|
|
- 2019: 76.9%
|
|
|
- 2020: 94.1%
|
|
|
|
|
|
## Limitations
|
|
|
|
|
|
- Trained on historical data only (2000-2020)
|
|
|
- May not perform well in unprecedented market conditions
|
|
|
- Requires proper risk management
|
|
|
- No consideration of transaction costs, slippage, or market impact
|
|
|
- Model predictions are probabilistic, not guaranteed
|
|
|
|
|
|
## Usage
|
|
|
|
|
|
### Prerequisites
|
|
|
```python
|
|
|
pip install joblib scikit-learn pandas numpy
|
|
|
```
|
|
|
|
|
|
### Loading the Model
|
|
|
```python
|
|
|
import joblib
|
|
|
import pandas as pd
|
|
|
from sklearn.preprocessing import StandardScaler
|
|
|
|
|
|
# Load model
|
|
|
model = joblib.load('trading_model.pkl')
|
|
|
|
|
|
# Load scalers (you need to recreate or save them)
|
|
|
# ... preprocessing code ...
|
|
|
|
|
|
# Prepare features
|
|
|
features = prepare_features(your_data)
|
|
|
prediction = model.predict(features)
|
|
|
probability = model.predict_proba(features)
|
|
|
```
|
|
|
|
|
|
### Features Required
|
|
|
The model expects 23 features in this order:
|
|
|
1. Close
|
|
|
2. High
|
|
|
3. Low
|
|
|
4. Open
|
|
|
5. Volume
|
|
|
6. SMA_20
|
|
|
7. SMA_50
|
|
|
8. EMA_12
|
|
|
9. EMA_26
|
|
|
10. RSI
|
|
|
11. MACD
|
|
|
12. MACD_signal
|
|
|
13. MACD_hist
|
|
|
14. BB_upper
|
|
|
15. BB_middle
|
|
|
16. BB_lower
|
|
|
17. FVG_Size
|
|
|
18. FVG_Type_Encoded
|
|
|
19. OB_Type_Encoded
|
|
|
20. Recovery_Type_Encoded
|
|
|
21. Close_lag1
|
|
|
22. Close_lag2
|
|
|
23. Close_lag3
|
|
|
|
|
|
## Training Details
|
|
|
|
|
|
- **Algorithm**: XGBoost Classifier
|
|
|
- **Hyperparameters**:
|
|
|
- n_estimators: 200
|
|
|
- max_depth: 7
|
|
|
- learning_rate: 0.2
|
|
|
- scale_pos_weight: 1.17 (for class balancing)
|
|
|
- **Cross-validation**: 3-fold
|
|
|
- **Optimization**: Grid search on hyperparameters
|
|
|
|
|
|
## SMC Strategy Elements
|
|
|
|
|
|
The model incorporates Smart Money Concepts:
|
|
|
- **Fair Value Gaps (FVG)**: Price imbalances between candles
|
|
|
- **Order Blocks (OB)**: Areas of significant buying/selling
|
|
|
- **Recovery Patterns**: Pullbacks in trending markets
|
|
|
|
|
|
## Upload to Hugging Face
|
|
|
|
|
|
To share this model on Hugging Face:
|
|
|
|
|
|
1. Create a Hugging Face account at https://huggingface.co/join
|
|
|
2. Generate an access token at https://huggingface.co/settings/tokens with "Write" permissions
|
|
|
3. Test your token: `python test_token.py YOUR_TOKEN`
|
|
|
4. Upload: `python upload_to_hf.py YOUR_TOKEN`
|
|
|
|
|
|
The script will upload:
|
|
|
- `trading_model.pkl` - The trained XGBoost model
|
|
|
- `README.md` - This model card with metadata
|
|
|
- All dataset files (CSV format)
|
|
|
|
|
|
## Citation
|
|
|
|
|
|
If you use this model in your research, please cite:
|
|
|
|
|
|
```
|
|
|
@misc{xauusd-trading-ai,
|
|
|
title={XAUUSD Trading AI Model with SMC Strategy},
|
|
|
author={AI Trading System},
|
|
|
year={2025},
|
|
|
url={https://huggingface.co/JonusNattapong/xauusd-trading-ai-smc}
|
|
|
}
|
|
|
```
|
|
|
|
|
|
### Academic Paper
|
|
|
For the complete academic research paper with methodology, results, and analysis:
|
|
|
|
|
|
**arXiv Paper**: [XAUUSD Trading AI: A Machine Learning Approach Using Smart Money Concepts](https://arxiv.org/abs/XXXX.XXXXX)
|
|
|
|
|
|
## License
|
|
|
|
|
|
This model is released under the MIT License. See LICENSE file for details.
|
|
|
|
|
|
## Contact
|
|
|
|
|
|
For questions or issues, please open an issue on the Hugging Face repository. |
