The selection of the Support Vector Machine (SVM) parameters is of great importance,
as it significantly influences the performance of the SVM algorithm. The Particle Swarm
Optimization (PSO) algorithm is ecient and widely used in solving many real world
problems. In this work, a method for finding optimal parameters based on particle swarm
optimization is used, with the objective of improving the learning and generalization
capacity of the SVM. The PSO-SVM is used in this work for the diagnosis of breast
cancer. The e↵ectiveness of the PSO-SVM algorithm was evaluated against the Wisconsin
Breast Cancer Dataset (WBCD), which is a database commonly used among researchers
using machine learning methods to diagnose breast cancer, and against the Early stage
diabetes risk prediction, which is a database for predicting early-stage diabetes risk, and
lastly in relation to the Sonar, Mines vs. Rock. The kernel function used is Radial Basis
Function (RBF). The results of the experiment demonstrate that the PSO-SVM algorithm
achieved a very considerable accuracy when compared to other studies that use only the
SVM algorithm for the diagnosis of breast cancer and for the prediction of diabetes risk.
The use of PSO proved to be ecient in the search for SVM parameters.
|
