Parkinson's Disease Recognition Using Gauss Map Based Chaotic Particle Swarm-Neural Network

Abstract

In detection of Parkinson's disease (PD), voice recordings are frequently appealed to reveal whether disease is available or not. The features extracted from these recordings are utilized as the input of classification methods. Herein, binary classification of features gains importance to accurately perform the PD detection. In this paper, we perform the classification of two well-known PD datasets including the features attained by recordings. Efficient hybrid classifiers are formed using the state-of-the-art optimization algorithms originated from particle swarm optimization (PSO). As an efficient classifier, neural network (NN) is determined as the main part of hybrid architecture. Sine map based chaotic PSO (SM-CPSO), dynamic weight PSO (DWPSO) and chaotic dynamic weight PSO (CDW-PSO) are considered to compare with Gauss map based CPSO (GMCPSO) on formation of hybrid classifiers and on classification of PD. For a detailed assessment, seven metrics (accuracy, AUC, sensitivity, specificity, g-mean, precision, f-measure) based comparison is realized, and 2-fold cross validation is handled to test the system. According to the results, GM-CPSO-NN achieves to remarkable performance among other hybrid methods and also outperforms to the recent literature studies. Consequently, a comprehensive study about PD recognition is realized, and a detailed comparison of hybrid NNs is presented on pattern classification. © 2019 IEEE.