GRU and LSTM Price Forecasting with Python and ONNX in MetaTrader

Summary

The article introduces machine learning for price prediction, then outlines a workflow using MetaTrader 5 hourly EUR/USD data, Pandas, scikit-learn, and neural networks. It explains basic decision trees and the GRU architecture, comparing its reset and update gates with the LSTM’s separate cell state and three gates. The practical aim is to train a price regression model in Python, export it to ONNX, and use it in a MetaTrader Expert Advisor. The text also mentions comparing GRU and LSTM models and using validation losses to help choose when to stop training.

The evidence is limited: it refers to results over January 2024 but provides no readable performance figures or detailed backtest. There is also an inconsistency: the described selected model is a GRU, while the shown layer examples are dense networks and an LSTM example, rather than a clear GRU implementation. The article suggests GRUs may train faster with similar results, but does not establish that claim for trading. Readers should treat its guidance as an implementation overview, not evidence of a profitable forecasting system.

Key ideas

• GRUs use reset and update gates to process sequential data with a simpler state structure than LSTMs.
• The proposed workflow uses historical MetaTrader 5 prices to train a regression model and export it for an Expert Advisor.
• The article describes a chronological train and test split, but does not provide a complete trading evaluation.
• The displayed neural network examples do not clearly match the article’s stated GRU model.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.