Discussing the article: "Neural Networks in Trading: An Agent with Layered Memory (Final Part)"

[MetaQuotes]

Check out the new article: Neural Networks in Trading: An Agent with Layered Memory (Final Part).

We continue our work on creating the FinMem framework, which uses layered memory approaches that mimic human cognitive processes. This allows the model not only to effectively process complex financial data but also to adapt to new signals, significantly improving the accuracy and effectiveness of investment decisions in dynamically changing markets.

The last two articles focused on the FinMem framework. In them, we implemented our interpretation of the approaches proposed by the framework authors using MQL5. We have now reached the most exciting stage: evaluating the effectiveness of the implemented solutions on real historical data.

It is important to emphasize that during the implementation we made significant modifications to the FinMem algorithms. Consequently, we assess only our implemented solution, not the original framework.

The model was trained on historical data for the EURUSD currency pair for 2023 using the H1 timeframe. Indicator settings analyzed by the model were left at their default values.

For the initial training phase we used a dataset formed during previous research. The implemented training algorithm, which generates "near-ideal" target actions for the Agent, allows training the model without updating the training dataset. However, to cover a broader range of account states, I would recommend adding periodic updates to the training dataset where possible.

After several training cycles we obtained a model that demonstrated stable profitability on both training and test data. Final testing was carried out on historical data for January 2024 with all other parameters unchanged. The test results are presented below.

Author: Dmitriy Gizlyk