Growing Neural Gas for Adaptive Clustering of Trading Data
Summary
This article explains Growing Neural Gas, an unsupervised clustering method whose network topology expands and changes as it learns the distribution of input vectors. Starting from two neurons, it repeatedly finds the closest and second-closest neurons, adjusts their weights and local error, updates or creates an edge, and removes connections that have aged past a limit. At intervals, it inserts a neuron near a high-error region. The article also describes linked lists as a way to manage the evolving neuron and edge sets in MQL5.
A later section outlines a utility-based variant that can remove neurons judged less useful, with the utility decay affecting how quickly the model adapts. The discussion presents the algorithm and implementation structure rather than a trading signal or empirical trading study. It notes that forgetting can help with slowly drifting input distributions but does not make the method responsive to rapid shifts. Stopping criteria and numerous parameters remain design choices, and the article supplies no trading performance comparison to establish an advantage over other clustering methods.
Key ideas
- Growing Neural Gas adapts both cluster representatives and the connections between them as data arrive.
- Neurons with large local error identify regions where inserting another representative may improve the model.
- Aging edges are removed to keep the network topology aligned with the evolving data distribution.
- Linked lists provide a practical way to manage neuron and edge collections that grow and shrink.
- Forgetting can support adaptation to slow distribution drift, but the article cautions that rapid changes remain difficult to track.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.