Machine learning-based classification of Parkinson's disease using acoustic features: Insights from multilingual...

Area of Science:

• Computational linguistics
• Neurology
• Machine learning

Background:

• Parkinson's disease (PD) prevalence is increasing, particularly in the elderly.
• Current PD diagnosis methods are time-consuming, costly, and not always accessible.
• Automated diagnostic tools are crucial for efficient and continuous PD monitoring.

Purpose of the Study:

• To develop and validate a robust machine learning model for automated Parkinson's disease diagnosis using speech analysis.
• To address the research gap in classifying speech patterns of Korean PD patients.
• To identify effective speech features for improved PD classification accuracy.

Main Methods:

• Utilized a voting-based machine learning approach for classifying speech patterns.
• Employed straightforward data preprocessing techniques.
• Incorporated the eGeMAPSv2 feature set and introduced novel features to enhance classification.

Main Results:

• Achieved an accuracy of 84.73% and an Area Under the ROC Curve (AUC) score of 92.18%.
• Demonstrated superior performance, especially with limited training data.
• Validated the model on a dataset of 100 Korean PD patients and 100 healthy controls.

Conclusions:

• The proposed model provides a practical and accurate solution for automated Parkinson's disease diagnosis.
• The developed system is suitable for integration into applications like smartphone-based diagnostic tools.
• Future work will focus on performance enhancement and exploring feature-PD relationships.