Hyper Parameters Selection by using Metaheuristic Algorithm for Improving Support Vector Regression

Abstract

Support vector regression (SVR) is one of the most accuracy machine learning techniques. However, the performance of this method depends on the selection of its hyperparameters. In most of the research, Grid search algorithm was used to select these hyperparameters and consider this algorithm among the best algorithms. In this paper, crow search algorithm is used to select the best combination of hyperparameters and then use them in the SVR method. Experimental results, obtained by running on three datasets, show that the crow search algorithm is performance of SVR was better when using the grid search algorithm in terms of prediction. In addition, it was demonstrated that the crow's search algorithm was better for its speed to obtain the best set of parameters. Alongside, By experimental results the crow search algorithm improving confirm the efficiency of the proposed algorithm in improving the prediction performance and computational time compared to other nature-inspired algorithms.