Abstract

Parkinson’s disease (PD) is a progressive neurological disorder that impairs the physical abilities of human beings such as speech, gait, and complex muscle-and-nerve actions. Early diagnosis of PD is important for alleviating the symptoms and reducing the overall socio-economic burden in healthcare. However, the key to the success is cost-effective and convenient telemedicine technology to overcome the challenges around the availability of expensive diagnoses and skilled clinicians. Almost 90% of PD patients suffer from a speech disorder and hence speech signals can be used as a non-invasive and cost-effective technique for diagnosis. Recent advances in machine learning and sensor technologies facilitate automated PD diagnosis for increased prediction accuracy and the capability to handle diverse biomarkers in an effective and timely manner. In establishing an effective predictive model for PD diagnosis, this study explores several popular machine learning classifiers in combination with feature set extraction, with and without Principal Component Analysis (PCA). Evaluating performance with a PD benchmark data set, our research discovered that Deep Neural Networks and Gradient Boosting Machines achieved 1st and 2nd places with an accuracy of 94.44% and 89.74%, with a weighted average f1-score of more than 93%. It is expected that this outcome will help clinicians achieve effective differentiation of the PD group from healthy controls based on voice data
Original languageEnglish
Title of host publicationArtificial Intelligence in Healthcare
EditorsTianhua Chen, Jenny Carter, Mufti Mahmud, Arjab Singh Khuman
PublisherSpringer Singapore
Publication statusAccepted/In press - Mar 2022

Publication series

NameBrain Informatics and Health
PublisherSpringer
ISSN (Print)2367-1742
ISSN (Electronic)2367-1750

Fingerprint

Dive into the research topics of 'Effective Diagnosis of Parkinson’s Disease using Machine Learning Techniques'. Together they form a unique fingerprint.

Cite this