Metaheurı́stica multiobjetivo para support vector machines con feature selection.
Loading...
Date
2024
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad de Concepción
Abstract
La selección de features es una área importante en el aprendizaje supervisado, especialmente al manejar grandes volúmenes de datos con muchas features. Support Vector Machines (SVM) ha demostrado ser un modelo sencillo y eficiente para trabajar en la selección de features gracias al uso regresores especı́ficos para cada feature, los cuales son anulados si no son relevantes. Por ello, esta memoria de tı́tulo propone una metaheurı́stica multiobjetivo para mejorar la selección de features aplicando SVM. La metaheurı́stica multiobjetivo se basa en el NSGA-II y busca optimizar tanto el rendimiento predictivo como la eficiencia computacional del modelo.
El enfoque propuesto es evaluado en diversos conjuntos de datos, considerando métricas de clasificación y tiempo de ejecución. Los resultados muestran que la metaheurı́stica presentada no solo reduce la dimensionalidad del problema sino que también mantiene o mejora la calidad de las predicciones.
Este trabajo contribuye al campo de la inteligencia artificial y la investigación de operaciones al demostrar que el uso de metaheurı́sticas pueden ser efectivas en la mejora de algoritmos de aprendizaje supervisado como SVM.
Feature selection is an important area in supervised learning, especially when handling large volumes of data with many features. Support Vector Machines (SVM) has proven to be a simple and efficient model to work on feature selection thanks to the use of feature-specific regressors, which can be overridden if they are not relevant. Therefore, this thesis proposes a multi-objective metaheuristic to improve feature selection using SVM. The multi-objective metaheuristic is based on NSGA-II and seeks to optimize both the predictive performance and the computational efficiency of the model. The proposed approach is evaluated on various datasets, considering classification metrics and runtime. The results show that the designed metaheuristic not only reduces the dimensionality of the problem but also maintains or improves the quality of the predictions. This work contributes to the field of artificial intelligence and operations research by demonstrating that the use of metaheuristics can be effective in improving supervised learning algorithms such as SVM.
Feature selection is an important area in supervised learning, especially when handling large volumes of data with many features. Support Vector Machines (SVM) has proven to be a simple and efficient model to work on feature selection thanks to the use of feature-specific regressors, which can be overridden if they are not relevant. Therefore, this thesis proposes a multi-objective metaheuristic to improve feature selection using SVM. The multi-objective metaheuristic is based on NSGA-II and seeks to optimize both the predictive performance and the computational efficiency of the model. The proposed approach is evaluated on various datasets, considering classification metrics and runtime. The results show that the designed metaheuristic not only reduces the dimensionality of the problem but also maintains or improves the quality of the predictions. This work contributes to the field of artificial intelligence and operations research by demonstrating that the use of metaheuristics can be effective in improving supervised learning algorithms such as SVM.
Description
Tesis presentada para optar al título de Ingeniero Civil Industrial
Keywords
Aprendizaje supervisado (aprendizaje de máquina), Estudio Supervisado (Aprendizaje de Máquina), Heurística