AI Trading Bot — Windows Native

An autonomous AI trading bot for MetaTrader 5 that uses machine learning (Logistic Regression, Random Forest, Gradient Boosting) to generate BUY/SELL signals for forex and gold pairs.

Supported Symbols

Symbol	Lot	SL (pips)	TP (pips)
EURUSD	0.10	50	100
XAUUSD	0.01	100	200
USDJPY	0.01	50	100
GBPUSD	0.01	50	100

Architecture

start.bat
    └── watchdog.py          Process manager (auto-restarts on crash)
            ├── real_bot.py  AI trading loop
            └── dashboard.py Streamlit web UI → http://localhost:8501

real_bot.py uses MetaTrader5 (native Windows library)
    → mt5.initialize() launches terminal64.exe automatically

Quick Start

1. Prerequisites

• MetaTrader 5 — install from your broker or metatrader5.com
• Python 3.10+ — install from python.org (check "Add to PATH")

2. Configure credentials

Edit config.py and set your MT5 account details:

MT5_LOGIN    = 12345678
MT5_PASSWORD = "your_password"
MT5_SERVER   = "YourBroker-Server"

3. Train models (first time only)

Open a terminal in the project folder:

python -m venv venv
venv\Scripts\pip install -r requirements.txt

:: Collect historical data (MT5 must be open and logged in)
venv\Scripts\python get_data.py

:: Train ML models for all 4 symbols
venv\Scripts\python train_model.py

This creates EURUSD_model.pkl, XAUUSD_model.pkl, USDJPY_model.pkl, GBPUSD_model.pkl.

4. Start the bot

start.bat

On first run start.bat automatically creates the venv and installs dependencies. The watchdog starts the bot and dashboard, then monitors both and restarts either if they crash.

To force-restart a running instance:

start.bat --force

Press Ctrl-C to stop everything cleanly.

5. Dashboard

Open http://localhost:8501 in a browser to see live signals, candlestick charts, account metrics, and trade history per symbol.

How It Works

1. Data collection (get_data.py) — fetches 10,000 M15 candles per symbol from MT5 and calculates RSI(14), MA20, MA50, and a 5-candle lookahead label.
2. Training (train_model.py) — tries Logistic Regression, Random Forest, and Gradient Boosting with 5-fold cross-validation; saves the best model per symbol as a .pkl file.
3. Trading loop (real_bot.py) — every 60 seconds:
   • Skips symbols with an open trade
   • Fetches the latest 100 candles and calculates indicators
   • Runs the ML model; if confidence ≥ 60% places a BUY or SELL order with SL/TP
   • Detects closed trades and sends a Windows desktop notification
4. Watchdog (watchdog.py) — monitors the bot and dashboard with exponential back-off restarts (30s → 60s → 120s … capped at 5 min).

Configuration

All settings are in config.py:

Setting	Default	Description
CONFIDENCE_THRESHOLD	0.60	Minimum model confidence to place a trade
LOOP_INTERVAL	60	Seconds between market checks
RESPECT_MARKET_HOURS	True	Skip trading outside Forex hours (Sun 22:00 – Fri 22:00 UTC)
MAGIC	123	MT5 magic number to identify bot trades
MAX_RECONNECT_ATTEMPTS	10	MT5 reconnect retries before exit

File Structure

Trading/
├── start.bat          Windows launcher
├── watchdog.py        Process manager
├── real_bot.py        AI trading loop
├── dashboard.py       Streamlit web dashboard
├── config.py          All settings and credentials
├── train_model.py     ML model training
├── get_data.py        Historical data collection
├── notifier.py        Windows desktop notifications (plyer)
├── requirements.txt   Python dependencies
├── *_model.pkl        Trained models (one per symbol)
├── *_data.csv         Training datasets (one per symbol)
└── logs/
    ├── bot.log        General bot activity
    ├── trades.log     All trade open/close events
    ├── watchdog.log   Watchdog activity
    └── dashboard.log  Streamlit output

Dependencies

MetaTrader5    Native Windows MT5 Python library
pandas         Data manipulation
numpy          Numerical operations
scikit-learn   ML models
joblib         Model serialisation
streamlit      Web dashboard
plotly         Interactive charts
plyer          Windows desktop notifications

Install: venv\Scripts\pip install -r requirements.txt