Thesis Diseño e implementación de un algoritmo genético para reducción de dimensionalidad para optimizar un clasificador objetivo
No Thumbnail Available
Date
2024
Authors
Journal Title
Journal ISSN
Volume Title
Program
Ingeniería Civil Informática
Departament
Campus
Campus Santiago San Joaquín
Abstract
La maldición de la dimensionalidad es un fenómeno que se manifiesta como un crecimiento exponencial de la complejidad computacional y la dificultad de aprendizaje a medida que aumenta el número de características en los datos utilizados en el campo del machine learning. Este trabajo propone como alternativa la aplicación de un algoritmo genético (AG) como método de selección de características para la reducción de atributos en modelos de clasificación, con el fin de mejorar el rendimiento general de las predicciones. La implementación de este algoritmo se llevó a cabo en el contexto del proyecto Human Place, utilizando redes neuronales y random forest como modelos de clasificación. Como parte de la validación, el algoritmo se aplicó a un segundo caso de estudio, para los datos de human activity recognition. Los resultados de esta investigación mostraron que el AG mejoró el rendimiento de las redes neuronales en el primer caso y fue más efectivo para el clasificador support vector machine (SVM) en el segundo caso.
The curse of dimensionality is a phenomenon characterized by an exponential increase in computational complexity and learning difficulty as the number of features in the data used in machine learning grows. This work proposes the use of a genetic algorithm (GA) as a feature selection method for attribute reduction in classification models, aiming to improve the overall performance of predictions. The implementation of this algorithm was carried out within the context of the Human Place project, using neural networks and random forest as classification models. As part of the validation, the algorithm was applied to a second case study involving human activity recognition dataset. The results of this research showed that the GA improved the performance of neural networks in the first case and was more effective for the support vector machine (SVM) classifier in the second case.
The curse of dimensionality is a phenomenon characterized by an exponential increase in computational complexity and learning difficulty as the number of features in the data used in machine learning grows. This work proposes the use of a genetic algorithm (GA) as a feature selection method for attribute reduction in classification models, aiming to improve the overall performance of predictions. The implementation of this algorithm was carried out within the context of the Human Place project, using neural networks and random forest as classification models. As part of the validation, the algorithm was applied to a second case study involving human activity recognition dataset. The results of this research showed that the GA improved the performance of neural networks in the first case and was more effective for the support vector machine (SVM) classifier in the second case.
Description
Keywords
Machine learning, Algoritmo genético, Maldición de dimensionalidad, Selección de características, Curse of dimensionality
