Kohonen Networks for Robust Parameter Selection and Time-Series Forecasting
Summary
The article presents practical uses of Kohonen self-organizing maps in algorithmic trading. For selecting expert-advisor settings, it recommends looking for broad plateaus of acceptable objective values rather than isolated backtest maxima, treating plateau width as a sign of robustness. It examines how genetic optimization can overpopulate profitable regions and miss nearby poor settings, and discusses fuller or hierarchical searches, pooling repeated runs under different criteria, and masking features to study parameters separately from economic results.
The second application forecasts a selected series from other market inputs. A silver forecast example uses currency and metals series and reports that accuracy sometimes reaches 60 percent during the cited test period. This is a limited demonstration: the author says forecasting depends on careful selection of the target, inputs, history depth, and tuning, and suggests external features or combined networks as possible improvements. Training within an indicator can block it, so scheduled map generation in an expert advisor is recommended.
Key ideas
- A broad plateau of acceptable optimization results may indicate more robust settings than a single peak.
- Genetic searches can under-sample weak parameter regions and hide losses adjacent to profitable settings.
- Repeated searches and feature masks offer ways to examine parameter and performance relationships in Kohonen maps.
- The forecasting example predicts silver using currency and metals data and reports occasional accuracy of 60 percent in its test period.
- Forecast quality depends on input preparation and extensive tuning, and the example does not establish general reliability.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.