Using Multi-objective Algorithms for Optimizing Support Vector Regression Parameters

Finding the perfect combination of Support Vector Regression (SVR) parameters to minimize two well-know regression metrics, Coefficient of Correlation (R2) and Root Mean Squared Error (RMSE), which are conflicting to each other, is a difficult task. To solve this problem, we propose four new regression models hybridized with multi-objective algorithms. We validated our algorithms using simulation data with and without noise, and real-world data sets. For each algorithm proposed, we analyzed the performance of the multi-objective algorithms and the regression results through convergence tests, descriptive statistics analyzes and hypothesis test. The results show that multi-objective algorithms are recommended for tunning the SVR parameters and finding feasible solutions for a given problem helping the decision maker to choose the best trade-off among these solutions.