Seleção de características para otimização de modelos de aprendizagem de máquina: uma abordagem baseada em algoritmos genéticos

Abstract

Predictive machine learning models rely on an appropriate set of attributes to perform classifications and predictions efficiently. However, the high dimensionality of data presents significant challenges, where the quality and relevance of features directly influence model performance, potentially leading to problems such as increased computational complexity and difficulty in finding relevant patterns, a phenomenon known as the "curse of dimensionality."This work investigates the application of genetic algorithms for feature selection in machine learning models, aiming to optimize their predictive performance by selecting only the relevant dimensions for the problem. Genetic algorithms (GAs) are metaheuristics inspired by the process of natural selection, employing operations such as selection, crossover, and mutation to evolve solutions across successive generations. These techniques are widely used to solve complex optimization problems, especially those involving high-dimensional search spaces. The study addresses the application of both the Conventional Genetic Algorithm (GA) and its random-key-based variant, BRKGA (Biased Random-Key Genetic Algorithm). The experiments demonstrated that both approaches are effective in reducing the dimensionality of data, improving performance metrics through the optimization of the F1-Score. The proposed methodology not only contributes to the development of more robust and scalable models but also provides greater interpretability and computational efficiency.