Selecting an optimal architecture of neural network using genetic algorithm
| Autor: | Sofia S. Emtseva, Aleksandr S. Gridin, Vladislav S. Fail, Jenny V. Domashova |
|---|---|
| Rok vydání: | 2021 |
| Předmět: |
education.field_of_study
Fitness function Artificial neural network Computer science business.industry Computer Science::Neural and Evolutionary Computation Crossover Population Software Multilayer perceptron Encoding (memory) Genetic algorithm General Earth and Planetary Sciences Artificial intelligence business education General Environmental Science |
| Zdroj: | BICA |
| ISSN: | 1877-0509 |
| Popis: | The article presents the results of applying a genetic algorithm to find the most optimal architecture of the neural network that would solve classification problem with minimal errors. The stages of the genetic algorithm are considered and the rule for encoding the parameters of the neural network is determined. A genetic algorithm for constructing the optimal architecture of a multilayer perceptron for solving classification problems has been developed. The algorithm independently creates a random population, evolves, creating new generations with more adapted individuals, i.e., neural networks with better architectures than previous generations. The paper describes the process of population formation, substantiates the choice of the fitness function and the method of selecting parents. Modifications of the crossover and mutation operators are proposed in order to ensure the operability of the algorithm on variable size individuals. A software tool that generates a neural network with the best parameters for solving classification problems has been developed on Python3 programming language. The architecture of a neural network for detecting fraudulent transactions has been built by using the developed software. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|