Automated Parkinson's disease detection and classification from speech data

Parkinson's disease is a progressive and neurodegenerative disorder that affects millions of individuals worldwide. This thesis aims to develop novel methods for the diagnosis of Parkinson's disease through the analysis of speech signals. The study demonstrates that specific acoustic features within speech signals can serve as biomarkers for the detection and classification of Parkinson's disease. The research utilized a sample group consisting of 28 healthy controls, 22 drug-naive Parkinson's patients (Med-off), and 30 medicated Parkinson's patients (Med-on), totaling 80 participants. The dataset was constructed by recording participants reading a predefined text in a quiet environment at the Neurology Department of Fırat University. From these recordings, a total of 19 acoustic features were extracted and analyzed. Various machine learning algorithms, including Support Vector Machines (SVM), Naive Bayes, Decision Trees, k-Nearest Neighbors (k-NN), and a neural network model specifically developed for this study, were employed. The performance of these models was evaluated using 10-fold cross-validation. The findings revealed significant differences in speech signals that distinguish Parkinson's patients from healthy individuals. These differences were shown to play a crucial role in facilitating early diagnosis of the disease. This study contributes a non-invasive, effective, and innovative approach to the diagnosis of Parkinson's disease, enriching the existing body of literature.