Por favor, use este identificador para citar o enlazar este ítem:
http://comimsa.repositorioinstitucional.mx/jspui/handle/1022/417| PSOin Ridge Regression to elimínate multícollinearity ¡n a PTA weldíng process | |
| JORGE SAUL RODRIGUEZ VAZQUEZ | |
| ROLANDO JAVIER PRAGA ALEJO | |
| Acceso Abierto | |
| Atribución-CompartirIgual | |
| MODELOS DE REGRESIÓN | |
| Nowadays industrial processes present variability in their processes making their control difficult. Linear regression is a statistical tool that can solve this problem, unfortunately if the process variables are highly related, the model obtained by Ordinary Least Squares (OLS) is not suitable to control or predict. This condition is called multicollinearity. Fortunately, there are statistical metrics capable of detecting linear dependence, such as the variance-covariance matrix, the VIF (Variance Inflation Factors) and the condition number. Ridge Regression (RR) is a method that eliminates the problem of multicollinearity. The basic idea of RR is generate a parameter of bias 𝑘 that counteracts the dependency between the variables. There are methods that provide the value of bias 𝑘 but affect model fit. For this reason, other alternatives should be choose. Therefore, in this article, a global optimization was performed applying the PSO metaheuristic on the parameter of bias 𝑘, to obtain the value that eliminates multicollinearity without affecting the fit of the model, taking the coefficient of determination 𝑅2 as an objective function. The optimization was applied to a case study, the results were contrasted against the Ridge Regression method, obtaining better results for the Particle Swarm Optimization algorithm proposed in this article. | |
| 2020-08 | |
| Capítulo de libro | |
| Inglés | |
| Estudiantes Investigadores | |
| ESTUDIOS INDUSTRIALES | |
| Versión publicada | |
| publishedVersion - Versión publicada | |
| Aparece en las colecciones: | ARTÍCULOS/ CAPÍTULOS DE LIBRO / MEMORIAS DE CONGRESO |
Cargar archivos:
| Fichero | Tamaño | Formato | |
|---|---|---|---|
| Publicacion ENC 2020.pdf | 642.02 kB | Adobe PDF | Visualizar/Abrir |