Using Particle Swarm Optimization Algorithm To Adjust The Gaussian Process Regression Parameters

الملخص

Gaussian Process Regression (GPR) is one of the most important techniques of automated learning, and has become a hot research subject in prediction tasks, which proposed by Williams and Rasmussen in 1996.

GPR can successfully analyze various data sets and generate predictions with high prediction accuracy. Nevertheless, the main challenge is the selection of GPR parameters. However, there is no generally structured way, yet.

In this study, we proposed new method in investigating the capability of GPR parameters using particle swarm optimization algorithm (PSO), and their quality was verified by applying them on both artificial and realistic data. This method was also compared with the analytical or experimental selection method (AS) using the measurements of prediction error explanation (MSE, RMSE, MAPE), in order to obtain the ideal method to select these parameters.

The results showed that the proposed method gives the best predictive accuracy when used to select parameters Gaussian Process Regression.