XAUUSD-LSTM

A Python-based LSTM model for forecasting XAU/USD (Gold vs USD) prices using historical 15-minute data. This project allows you to train an LSTM neural network, evaluate its predictions, and generate future price forecasts.

---

Features

• Train an LSTM model on historical gold prices.
• Automatic preprocessing and resampling of data.
• Customizable sequence length, batch size, epochs, and validation split.
• Save best model checkpoint and final trained model.
• Generate recursive multi-step forecasts.
• Plot validation predictions and historical + forecast trends.
• Support for CSV input and output for easy integration.

---

Dataset

XAUUSD 15 minutes (2024 - 2025)

Date;Open;High;Low;Close;Volume
2004.06.11 07:15;384;384.3;383.8;384.3;12
2004.06.11 07:30;383.8;384.3;383.6;383.8;12
2004.06.11 07:45;383.3;383.8;383.3;383.8;20
...

https://www.kaggle.com/datasets/novandraanugrah/xauusd-gold-price-historical-data-2004-2024

---

Running

python train_xau_lstm.py --data "XAU_15m_data.csv" --datetime_col "Date" --target "Close" --seq_len 96 --epochs 5 --batch_size 64 --resample 15T

---

Requirements

• Python 3.9+
• TensorFlow 2.x
• NumPy
• pandas
• scikit-learn
• matplotlib

Install dependencies via:

pip install -r requirements.txt

(You can create requirements.txt with the following:)

tensorflow
numpy
pandas
scikit-learn
matplotlib

---

Usage

Command-Line Arguments

python train_xau_lstm.py --data "XAU_15m_data.csv" --datetime_col "Date" --target "Close" \
--seq_len 96 --epochs 30 --batch_size 64 --resample 15min --forecast_steps 16

Arguments:

Argument	Description	Default
--data	Path to the CSV file containing historical data	XAU_15m_data.csv
--datetime_col	Name of the datetime column in CSV (auto-detected if not provided)	None
--target	Target column for prediction	Close
--resample	Resample frequency (15min, H, D, etc.)	min
--seq_len	Sequence length for LSTM input	96
--val_split	Fraction of data for validation	0.2
--epochs	Number of training epochs	30
--batch_size	Training batch size	64
--patience	Early stopping patience (number of epochs)	6
--dropout	Dropout rate in LSTM layers	0.2
--model_path	Path to save the best model checkpoint	xau_lstm_best.keras
--final_model	Path to save the final trained model	xau_lstm_final.keras
--plot_path	Path to save validation prediction plot	prediction_plot.png
--forecast_steps	Number of future steps to forecast	16
--forecast_out	CSV output path for forecast results	forecast_next.csv

---

Output

1. Model files:

   • xau_lstm_best.keras: Best model checkpoint during training.
   • xau_lstm_final.keras: Final trained model.

2. Forecast CSV: Example forecast_next.csv:

,forecast
2025-09-30 19:45:00,3280.56
2025-09-30 20:00:00,3308.98
2025-09-30 20:15:00,3337.23
...
2025-09-30 23:30:00,3529.38

3. Plots:

   • prediction_plot.png: Validation actual vs predicted prices.
   • historical_forecast_plot.png: Historical prices + future forecast.

---

Example Workflow

# Train and forecast
python train_xau_lstm.py --data "XAU_15m_data.csv" --datetime_col "Date" --target "Close" \
--seq_len 96 --epochs 5 --batch_size 32 --resample 15min --forecast_steps 16

• Model trains on historical XAU/USD prices.
• Validation predictions are plotted.
• Next 16 steps forecast is saved to forecast_next.csv.

---

Notes

• Make sure your CSV uses ; as a separator.
• resample should match your data frequency (15min, 1H, etc.).
• For faster testing, reduce seq_len, epochs, or use a subset of the data.

---

License

This project is licensed under the MIT License.

---

Author

Max Base

GitHub: https://github.com/BaseMax/XAUUSD-LSTM

Copyright 2025, Max Base