AlphaFlow ML & DL Trading Bot Project

Multi-strategy MT5 research lab for ML/DL/time-series trading: data → modeling → backtests → tuning → prototype execution.

A comprehensive machine learning and deep learning trading framework that covers the entire workflow:

1. Data loading from MetaTrader 5
2. Feature engineering (technical indicators, custom features, labeling)
3. Model training (RandomForest, XGBoost, LightGBM, deep learning models, etc.)
4. Hyperparameter tuning (RandomizedSearchCV, GridSearchCV or Optuna)
5. Time-based / walk-forward cross-validation
6. Backtesting (VectorBT or simple custom code)
7. Live trading integration with MetaTrader 5

Supported Strategies:

• Regression on next-bar returns
• Multi-bar classification
• Double-barrier labeling (López de Prado style)
• Regime detection (simple up/down/sideways approach)
• Volatility-based labeling
• Momentum Strategy
• Pairs Trading (Cointegration)
• Pairs Trading (Clustering)

This project provides a flexible template for you to create and add your own custom labeling functions or feature engineering steps, allowing you to experiment with new ideas and strategies.

Table of Contents

1. Features
2. Repository Structure
3. Setup & Installation
4. Usage
   • Backtesting Notebooks
   • Live Trading Scripts
5. Key Modules
6. Extending the Project
7. Disclaimer
8. License

Features

• MetaTrader 5 data retrieval (data_loader.py)
• TA library for feature engineering (ta.add_all_ta_features)
• Multiple labeling methods: next-bar, multi-bar, double-barrier, regime detection, volatility-based etc.
• Time-based or walk-forward cross-validation to avoid data leakage
• RandomizedSearchCV or GridSearchCV for hyperparameter tuning
• VectorBT or custom backtesting scripts for performance evaluation
• Live trading scripts with real-time MetaTrader 5 order sending

Repository Structure

# AlphaFlow ML & DL Trading Bot Repository Structure

AlphaFlow-MT5-ML-DL-Trading-Lab/
├── data/
│   ├── data_loader.py              # MetaTrader 5 data retrieval
│
├── features/
│   ├── feature_engineering.py      # Technical indicators, stationarity checks, custom features
│   ├── labeling_schemes.py         # Labeling methods: next-bar, multi-bar, double-barrier, regime detection, volatility-based
│
├── models/
│   ├── model_training.py           # Model selection, hyperparameter tuning (Optuna, GridSearchCV)
│   ├── saved_models/               # Folder for saved model pipelines (.pkl, .joblib)
│
├── backtests/
│   ├── simple_backtest.py          # Simple event-driven backtest logic
│   ├── vectorbt_backtest.py        # VectorBT-based backtesting template
│
├── live_trading/
│   ├── ... (live trading scripts)
│
├── notebooks/
│   ├── exploratory/
│   │   ├── eda_visualization.ipynb
│   │   ├── integrated_pipeline_pair_trading.ipynb
│   │   └── kalman_filters_solution.ipynb
│   ├── strategies/
│   │   ├── ml/
│   │   │   ├── backtesting/
│   │   │   │   ├── regression_returns.ipynb
│   │   │   │   ├── double_barrier_labeling.ipynb
│   │   │   │   ├── multi_bar_classification.ipynb
│   │   │   │   ├── multi_bar_classification_multisymbol.ipynb
│   │   │   │   ├── multi_bar_classification_multisymbol_core_features.ipynb
│   │   │   │   ├── multi_bar_classification_multisymbol_core_features_stocks.ipynb
│   │   │   │   ├── regime_detection.ipynb
│   │   │   │   ├── momentum_strategy.ipynb
│   │   │   │   ├── pairs_trading_cointegration.ipynb
│   │   │   │   └── pairs_trading_clustering.ipynb
│   │   │   └── live_trading/
│   │   │       ├── regression_returns.ipynb
│   │   │       ├── double_barrier_labeling.ipynb
│   │   │       ├── multi_bar_classification.ipynb
│   │   │       ├── multi_bar_classification_multisymbol_core_features.ipynb
│   │   │       ├── regime_detection.ipynb
│   │   │       ├── pairs_trading_cointegration.ipynb
│   │   │       └── pairs_trading_clustering.ipynb
│   │   └── dl/
│   │       ├── backtesting/
│   │       │   ├── regression_returns.ipynb
│   │       │   └── multi_bar_classification_multisymbol.ipynb
│   │       └── live_trading/
│   │           └── regression_returns.ipynb
│   └── time_series/
│       └── arima_sarima_var_lstm.ipynb
│
├── pyproject.toml
├── uv.lock
├── README.md

Setup & Installation

1. Clone this repository:

git clone https://github.com/maghdam/AlphaFlow-MT5-ML-DL-Trading-Lab.git
cd AlphaFlow-MT5-ML-DL-Trading-Lab

2. Create and activate a Python environment (conda or venv):

conda create -n ml_trading python>=3.9
conda activate ml_trading

3. Install dependencies with uv:

pip install uv
uv pip sync pyproject.toml

• Make sure you have MetaTrader5 installed IC Markets MT5.

4. (Optional) Install Jupyter Notebook:

pip install jupyter

Usage

Backtesting Notebooks

1. Navigate to ml_notebooks/ or dl_notebooks/, pick a relevant file (e.g., 02_backtests_multi_bar_classification.ipynb), and run it:

   jupyter notebook

2. Inside the notebook, you can see how we do:
   • Feature engineering
   • Labeling
   • Walk-forward splits
   • Train & tune
   • VectorBT or custom backtesting

Live Trading Scripts

1. Navigate to ml_notebooks/ or dl_notebooks/, pick a relevant live trading file (e.g., 2_live_trading_multi_bar_classification.ipynb), or go to live_trading/ folder and pick the script for your labeling approach:
   • regression_returns.py
   • multi_bar.py
   • double_barrier.py
   • regime_detection.py
2. Adjust MetaTrader 5 credentials (login, server, password) in the script.
3. Run from terminal:

   python live_trading/multi_bar.py.py

4. The script will:
   • Load the pipeline (e.g., best_rf_mb_pipeline.pkl)
   • Fetch new bars from MetaTrader 5
   • Predict SHIFTED classes [0, 1, 2] => SHIFT back to [-1, 0, +1]
   • Place orders if signals = ±1

Key Modules

• data/data_loader.py: Connects to MetaTrader 5, fetches bars with copy_rates_from_pos.
• features/feature_engineering.py: Uses the TA library and additional custom features (spreads, autocorrelation, etc.).
• features/labeling_schemes.py:
  • calculate_future_return(...)
  • create_labels_multi_bar(...)
  • create_labels_double_barrier(...)
  • create_labels_regime_detection(...)
  • create_labels_volatility(...)
• models/model_training.py:
  • select_features_rf_reg(...)
  • Time-based splits, random/grid search for hyperparams.
• backtests/:
  • simple_backtest.py or vectorbt_backtest.py
• live_trading/:
  • Each script loads a pipeline (.pkl), connects to MT5, and places trades based on predictions.

Extending the Project

• Add your own label: Create a new function in features/labeling_schemes.py (e.g. create_labels_custom(...) that returns a new column with [-1, 0, +1] (or your custom classes)).
• Add your own features: Implement them in features/feature_engineering.py or create a new file.
• Train a new model: Adapt models/model_training.py or your notebooks to handle new classifiers/regressors.
• Explore new backtest approaches: Either integrate with vectorbt in a notebook or write a custom .py in backtests/.

Disclaimer

We share this code for learning and development/research purposes only. Nothing herein constitutes financial advice or a recommendation to trade real money. Trading involves substantial risk. Always do your own due diligence, consult professionals, and only risk capital you can afford to lose.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Backtest Results - US30 - H4

Loaded best classification model from 'best_rf_mb_pipeline.pkl'

Out-of-Sample Accuracy: 0.5439

Running Full Backtest on the Last 5000 Bars...

Full Backtest Results:
Accuracy=0.54, Return=0.30%, Sharpe=1.21
Start                               2021-12-01 16:00:00
End                                 2025-02-28 00:00:00
Period                                832 days 12:00:00
Start Value                                     10000.0
End Value                                  12971.402323
Total Return [%]                              29.714023
Benchmark Return [%]                          24.515943
Max Gross Exposure [%]                            100.0
Total Fees Paid                              236.291366
Max Drawdown [%]                              13.645737
Max Drawdown Duration                 295 days 04:00:00
Total Trades                                         56
Total Closed Trades                                  56
Total Open Trades                                     0
Open Trade PnL                                      0.0
Win Rate [%]                                  60.714286
Best Trade [%]                                13.156291
Worst Trade [%]                               -3.463323
Avg Winning Trade [%]                          1.517335
Avg Losing Trade [%]                          -1.094661
Avg Winning Trade Duration    7 days 03:03:31.764705882
Avg Losing Trade Duration               1 days 00:00:00
Profit Factor                                  2.150809
Expectancy                                    53.060756
Sharpe Ratio                                    1.20547
Calmar Ratio                                   0.885441
Omega Ratio                                    1.156419
Sortino Ratio                                  1.793852
dtype: object