Forex Market Prediction using Deep Learning

Overview

This project is an experiment to apply deep learning techniques to predict future market prices in the Forex market. The goal is to leverage advanced time series forecasting models to enhance predictive accuracy and inform trading strategies.

Installation and Setup

To run this project, follow the steps below:

1. Install Python and Create a Virtual Environment

Ensure you have Python 3.8 or later installed. You can check your version by running:

python --version

Create a Virtual Environment

1. Navigate to the project folder:

   cd path/to/project

2. Create a virtual environment:

   python -m venv forex_env

3. Activate the virtual environment:

   • Windows:

     forex_env\Scripts\activate

   • Mac/Linux:

     source forex_env/bin/activate

2. Install Dependencies

Once inside the virtual environment, install the required packages:

pip install -r requirements.txt

3. Download Data

The dataset for this project includes historical forex market prices and can be obtained from relevant data sources.

For benchmark datasets, you can refer to:

• Autoformer Repository
• Informer Repository

Ensure your dataset is placed inside the data/ folder before proceeding.

Changing the Dataset

To change the dataset and include a new dataset, you can run the data_downloading_and_feature_extraction0.ipynb notebook and modify the parameters within the notebook accordingly.

Running the Experiment

To execute the experiment, follow these steps:

1. Open Jupyter Notebook:

   jupyter notebook

2. Navigate to the project folder and open fedformer6.ipynb.

3. Run the notebook cells sequentially to execute the Forex market prediction script.

Model Architecture

This model utilizes an advanced time series forecasting transformer-based architecture. Below is the structure of the model:

Figure 1. Model architecture used for time-series forecasting

The model is designed to efficiently process long-term dependencies in time series data by leveraging frequency-enhanced decomposition techniques.

Experiment Execution

Once the setup is complete:

• Open the Jupyter notebook (fedformer6.ipynb).
• Execute all cells sequentially.
• The model will train on the dataset and provide market price predictions.

Citation

If this experiment is useful to you, consider referring to related work:

@inproceedings{zhou2022fedformer,
  title={Frequency enhanced decomposed transformer for long-term series forecasting},
  author={Zhou, Tian and Ma, Ziqing and Wen, Qingsong and Wang, Xue and Sun, Liang and Jin, Rong},
  booktitle={Proc. 39th International Conference on Machine Learning (ICML 2022)},
  location = {Baltimore, Maryland},
  year={2022}
}

Further Reading

For further understanding of transformers in time-series forecasting:

• Qingsong Wen, Tian Zhou, et al. "Transformers in time series: A survey." arXiv preprint.

Contact

For questions regarding this experiment, please reach out to the project contributors or refer to the linked repositories:

• Autoformer Repository
• Informer Repository
• Multivariate Time Series Data

---

This document provides a comprehensive guide to setting up and running the experiment. The model is built to offer insights into market trends using advanced deep learning techniques.