GMDH Neural Network MATLAB Implementation
GMDH-type Polynomial Neural Network A MATLAB implementation of Group Method of Data Handling (GMDH) for building polynomial neural networks. The algorithm iteratively constructs network layers where the exact architecture (connections) and size (number of layers) are determined automatically during training through evaluation criteria. Validation can use additional test data with regularity criteria or explicit complexity control via information criteria (AIC) or minimum description length (MDL). Key parameters include: number of inputs per neuron, maximum polynomial terms in neurons, cross-layer connections allowing inputs from both previous layers and original variables, and configurable neuron reduction in subsequent layers for optimal network complexity.