Predicting Parkinson's disease based on transformed speech signal by DWT with two methods for features selection: GA-AdaBoost and BPSO

In the medical domain, the cruelest phase is the diagnosis phase. Regarding the phase of diagnosis of the disease as Socrates said “understanding the questions is the half of the answer”, which means when we can make an accurate diagnosis that will be considered as reaching half of the healing process. This article proposes an accurate system for Parkinson's disease detection. Hence, we will interpret the efficacity of the wrapper methods Genetic Algorithm (GA), Binary Particle Swarm Optimization (BPSO), and machine learning algorithms in the diagnosis of Parkinson's disease the second most common neurologic disease after Alzheimer's disease. Based on the speech signal processing of patients uttering vowels by using the discrete wavelet transform DWT. Then extract the features matrix that contains Linear Predictive Coding (LPC), Zero-Crossing Rate (ZCR), Mel Frequency Cepstral Coefficient (MFCC), and wavelet Shannon entropy. Furthermore, a hybrid method applied the AdaBoost, KNN classifier, and the genetic algorithm to reduce the features and enhance the accuracy of detection. First, the genetic algorithm occurs in the selection of the features and the creation of a new population then applied a combination of KNN classifier and AdaBoost to validate the accuracy of the proposed system. A second time we use the BPSO to select the features. The best accuracy obtained was 100% by both of the models.