Parkinson’s Disease Detection Using Voice Features and Machine Learning Algorithms

This paper investigates the noninvasive screening for early signs of Parkinson’s disease from voice signals using machine learning algorithms. It considers 752 audio features extracted from phonation of the sustained vowel ‘/a/’ sound. Two machine learning algorithms, namely k-nearest neighbors (kNNs) and support vector machines (SVMs), are modeled to identify their effectiveness for the classification task. The results showed that cubic SVM and fine kNN algorithms could achieve 100% accuracy. However, the optimized kNN outperformed the optimized Gaussian SVM in terms of accuracy, detection speed, training time, and misclassification cost.