Suggested Method to Adjust the Parameters of SVM Regression using Particle Swarm Optimization

Abstract

The Gaussian Process Regression (GPR) was first introduced as a regression tool in the field of machine learning by Rasmussen and Williams in 1996, where they described the optimization of parameters in the covariance function, which was inspired by the use of the Gaussian process with neural networks.

This research aims primarily to study and improve the performance of the Gaussian process regression by reaching the best method for adjusting the values of the coefficients of this technique.

In this research, a new method war proposed for testing the adjustment of the values of the Gaussian process regression coefficients grounded on the Particle Swarm Optimization (PSO) algorithm, with the aim of reaching the best method for choosing the values of these coefficients.

In order to apply the two methods, a script was written in a statistical programming language (R), which performs many data analysis by using Packages.

Addicts is an important social aspects and dangerous phenomenon in population because its aim is to destroy the minds and bodies of young people and at the same time destroy them, therefore the phenomenon of drug addiction has become one of the serious problems that preoccupy officials in all parts of the world, especially our Islamic world, and day after day the danger of addiction is exacerbated, because it increases every day with the decrease in the age of addiction.

Realistic data was collected from the Ibn-Rushed Psychiatric Hospital in Baghdad , involved 196 males from July 2023 to February 2024 involved 196 males, with addiction durations with varying degrees of methamphetamine (METH) addiction, which represent the duration and  four test measure which is commonly used to help diagnose liver damage or disease   which are  Alanine aminotransferase test(ALT)   , Aspartate aminotransferase test  ( AST) ,Alkaline Phosphatase ( ALP) , and Gamma – glut amyl Transfers(  GGT).

The following measures ( MSE, RMSE, MAPE) are conceded the best measures for comparing different prediction models built using the same training dataset ( Hyndman and Koehler, 2006)..