The effectiveness of the GMDH neural network type algorithms and inductive modeling technologies developed on their basis, is confirmed by solving numerous real-world modeling problems in economics, ecology, and technology. However, today there are many problems of such complexity that to fully solve them it is not enough the capabilities of any particular method of computational intelligence, including GMDH.
Therefore, the idea of combining the GMDH neural network with the genetic method of computational intelligence in order to design more efficient hybrid tools is put forward. The originality of this idea is that the neurons of this neural network become so-called "active" through the use of combinatorial-genetic method to optimize the structure of the neuron. This will make it possible to form the neural network architecture and each neuron at the same time, which will mean building a so-called "neural network with active neurons", which will have a much higher ability to tune to a specific data sample compared to known neural network architectures.
The following main tasks will be performed during the project implementation:
1. Research of advantages and disadvantages of existing GMDH neural network architectures for inductive model construction;
2. Development of a new class of GMDH neural networks with active neurons;
3. Software implementation of the developed neural network of GMDH and research of its efficiency on test and real problems of different nature.
The applied significance of this topic is that the product of development will be a new software tool for data mining, modeling and forecasting of complex processes.
The final products of this theme will be: a new neural network GMDH with active neurons; appropriate software for building models of complex objects and processes; results of solving test and real problems.