Seeding Initial Population, in Genetic Algorithm for Features Selection
| Autor: | Charly Clairmont, Nicoleta Rogovschi, Nistor Grozavu, Marc Chevallier, Faouzi Boufarès |
|---|---|
| Rok vydání: | 2021 |
| Předmět: |
education.field_of_study
Computer science Population Bayesian network 020206 networking & telecommunications Feature selection 02 engineering and technology Random forest Convergence (routing) Classifier (linguistics) Genetic algorithm 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing education Algorithm Selection (genetic algorithm) |
| Zdroj: | Advances in Intelligent Systems and Computing ISBN: 9783030736880 SoCPaR |
| DOI: | 10.1007/978-3-030-73689-7_55 |
| Popis: | The feature selection process is a difficult task that can be tackled by various algorithms. Our work uses a subclass of metaheuristic algorithms called genetic algorithms (GA) to select the best subset of features that has given, for a machine learning algorithm, the best results (based on accuracy). GA are easy to implement and understand, and their results are readily explainable. However, they don’t ensure to find the absolute best solution for a given problem, but only the best solution found. In order to improve the performance of GA, we introduce two seeding methods for the initial population of the GA that rely on the use of a Random Forest algorithm. The two methods are applied on two different GA using Bayesian networks as classifier to evaluate accuracy. The tests are done on five data-sets, and the two methods are compared to other dimensional reduction techniques. Our results show a better convergence of the genetic algorithms when they are seeded. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|