Building a CNN Classifier for EURUSD Direction from OHLC Windows

Summary

The article outlines a computer vision workflow for classifying EURUSD price direction from OHLC time series. It retrieves hourly data from MetaTrader 5, forms rolling windows, scales each window, and labels it according to whether the closing price is higher after a forecast interval. A one-dimensional convolutional network with normalization, pooling, dropout, and dense layers is trained with early stopping.

It also proposes inspecting intermediate feature maps to visualize what the model represents. The described split is shuffled into training and validation subsets, and the article gives no reported predictive metrics, trading simulation, or evidence of out-of-sample profitability. The presentation therefore serves as an implementation outline, not proof that the model finds reliable signals. Its preprocessing and evaluation choices would need scrutiny before trading use, particularly the temporal split and how scaling relates to future observations.

Key ideas

• The workflow maps rolling OHLC windows to a binary future-direction label.
• A one-dimensional CNN extracts features from normalized sequences of hourly bars.
• Early stopping monitors validation loss and restores the best model weights.
• Intermediate feature maps are exposed for visualization of learned representations.
• The article reports no trading results or predictive metrics to establish model usefulness.

Tags

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