Using Volume Features, Clustering, and LSTMs to Forecast Price Moves

Summary

The article outlines a machine learning workflow that combines volume analysis with price features to predict market movements. It proposes rolling volume averages and ratios, price and volume momentum, volume volatility, and price-volume correlation. Isolation Forest is used to flag unusual volume observations, while K-means groups observations using volume, its relative level, and volatility. The author describes a simple LSTM model trained through Python and MetaTrader 5, and suggests that combining anomaly flags with clusters may add context to forecasts.

The claimed application is hourly data for Russian equities, with historical Sberbank testing and visualized results mentioned, though the supplied text does not give detailed performance statistics or enough validation design to assess robustness. The author reports that a more complex network overfit and favors a simpler architecture, while acknowledging that feature choice and model settings need further work. The reported patterns and effectiveness are exploratory claims; they do not establish durable predictive power or live trading profitability.

Key ideas

• Volume ratios, rolling averages, momentum, volatility, and price-volume correlation form the proposed feature set.
• Isolation Forest flags atypical volume, while K-means groups observations using volume and volatility features.
• The article proposes an LSTM to turn these features into price movement forecasts.
• The author reports possible interactions among anomalies, clusters, and subsequent price movement, especially near session openings.
• A complex network reportedly overfit, and the historical testing claims lack detailed metrics in the supplied text.

Tags

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