PARAMETER SEARCH SPACE ANALYSIS FOR SUPPORT VECTOR REGRESSION (SVR)

Support Vector Regression (SVR) is a powerful method that has been successfully applied to many regression problems. However, the effectiveness of SVR depends on the proper selection of its parameters. For this reason, this paper analyzed the characteristics of the SVR parameter search space for the Gaussian kernel function. Three datasets were studied using grid search. The experimental results revealed the existence of a good region where optimal parameter solutions could be identified. These solutions enabled SVR to achieve good performance, accurately predicting test data with a high correlation coefficient. Although further in-depth analyses are needed, identifying this optimal region contributes to new research, especially in developing more effective parameter selection methods that focus on this specific zone. Such approaches can optimize the search process and reduce processing time by avoiding the evaluation of parameters in regions prone to underfitting or overfitting.