Training a Stacked RBM Deep Network for EURUSD Signal Classification

Summary

The article describes building a deep belief network initialized with stacked restricted Boltzmann machines, then fine-tuning it with labeled data. For EURUSD on a 30-minute timeframe, it constructs inputs from technical indicators and predicts directional labels derived from a ZigZag signal. The preparation workflow includes removing missing values, excluding highly correlated predictors, and selecting variables by global, local, and class-specific importance. It also discusses splitting data into training, testing, and validation sets and balancing classes.

Model evaluation includes prediction metrics and post-processing through calibration, Markov-chain smoothing, and adjustment against a theoretical balance curve. An Expert Advisor periodically retrains the model during operation and monitors its results. The reported conclusion is that this RBM-initialized network trained less effectively than a prior stacked-autoencoder model, though it trained quickly and exposed many tuning options. The article gives no detailed numerical performance evidence in the supplied text, and its ZigZag-derived target and specific historical sample limit how broadly its conclusions can be applied.

Key ideas

• The network is pretrained layer by layer with RBMs before supervised fine-tuning.
• The EURUSD classifier uses technical-indicator inputs and ZigZag-derived directional labels.
• Data preparation includes missing-value removal, correlation filtering, and predictor-importance selection.
• Calibration and Markov-chain smoothing are presented as ways to refine model predictions.
• The Expert Advisor periodically retrains while monitoring its live results.

Tags

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