Eventos Anais de eventos
COBEM 2017
24th ABCM International Congress of Mechanical Engineering
Evolutionary Support Vector Regression Approach Applied to Backstroke Start Performance Modelling
Submission Author:
Leandro dos Santos Coelho , PR , Brazil
Co-Authors:
Karla de Jesus, Carlos Eduardo Klein, Leandro dos Santos Coelho, Kelly de Jesus, Viviana Mariani, Leandro Machado, Mário Vaz, ricardo fernandes, J. Paulo Vilas-Boas
Presenter: Carlos Eduardo Klein
doi://10.26678/ABCM.COBEM2017.COB17-1264
Abstract
Support Vector Machine (SVM) is one of the fastest growing methods of machine learning due to its good generalization ability and good convergence performance. SVM is a maximum margin model, which is based on structural risk minimization rather than empirical risk minimization. Originally, SVM developed for solving the classification problems but latter, Support Vector Regression (SVR) evolved from the SVM for doing regression tasks. A SVR model is, in essence, a machine learning method of non-parametric estimation especially aiming at samples with limited sizes. The principle of structural risk minimization makes SVR to have stronger generalization ability. Despite their advantages, SVR models require an accurate selection of the configuration parameters in order to achieve good generalization performance. To overcome this limitation, a hyperparameter selection method based on differential evolution (DE) optimizer was developed. The aim of this study is to develop an evolutionary SVR (E-SVR) approach based on DE selection method to model the backstroke start performance. SVR and the proposed E-SVR approach were applied and compared with other regression methods to predict 5 m backstroke start time using kinematic and kinetic variables and to determine the accuracy of the mean absolute percentage error. The results presented an excellent performance in terms of the prediction errors of both SVR and E-SVR approaches.
Keywords
Support vector regression, Nonlinear regression, Kinematics, Kinetics, Competitive swimming, Evolutionary Algorithm, Differential Evolution
