Optimization of support vector machines by meta heuristic methods and applying on parkinson's disease dataset

Parkinson's disease occurs as a result of the decrease and insufficiency of dopamine cells over the time. This decrease varies with age. Considering the fact that the world population is getting older, this disease may increase in the coming years. Although there is no definitive diagnosis method for this disease, the patient is kept under follow-up for a long time and then parkinson's disease can be diagnosed. In this thesis study, classification was made with Support Vector Machine (SVM) to detect parkinson's patients. The selection of parameters in SVM is very important. Incorrect parameter selection affects the classification success of SVM. Therefore, meta heuristic methods have been used extensively in the literature in recent years and successful results have been obtained. In this study, SVM optimization models were created using Bat Algorithm (BA), Improved Chaotic Particle Swarm Optimization (ICPSO), Improved Genetic Algorithm (IGA) and Harmony Search Algorithm (HSA). Experiments in the study were conducted on two different parkinson's disease datasets taken from the UCI (Machine Learning Repository of University of California at Irvine) database and these datasets were named as Parkinson 1 and Parkinson 2. Obtained results were compared using accuracy, sensitivity, specificity, precision and F1 score evaluation criteria. Among DVM kernel functions, RBF, linear and polynomial kernel functions were used. In general, RBF kernel function has the most successful performance in both datasets. The best accuracy results in Parkinson 1 and Parkinson 2 datasets belong to ICPSO-SVM model with 88.75% and BA-DVM model 95.422%, respectively. Literature comparisons have shown that the proposed models are better than some studies and can compete with some studies.