Table 1.
Definitions of vocal features.
Table 2.
Statistical parameters of vocal features.
Fig 1.
Number of positive (above threshold) and negative instances (below threshold) with respect to determined UPDRS threshold.
Fig 2.
(left) Test set classification accuracies and (right) Matthew's correlation coefficients obtained with k-NN classifier under various UPDRS threshold values.
Fig 3.
(left) Test set classification accuracies and (right) Matthew's correlation coefficients obtained with SVM classifier under various UPDRS threshold value.
Fig 4.
(left) Test set classification accuracies and (right) Matthew's correlation coefficients obtained with ELM classifier under various UPDRS threshold values.
Fig 5.
A summary of results obtained with the best settings of classifiers (left) Matthew's correlation coefficients of the classifiers obtained with their best settings (right) ROC space of the classifiers obtained when UPDRS threshold is set to 15.
Fig 6.
Scatter of PD data on the first three principal components with UPDRS threshold value of (top) (left) 15 (right) 20 (bottom) (left) 25 (right) 30.
Fig 7.
Absolute difference between the ratio of the number of patients whose UPDRS is below the corresponding threshold to the number of all patients in cluster 1 and cluster 2.
Table 3.
Ranking of the vocal features based on their mutual information with UPDRS level discretized according to the determined optimal threshold that can be discriminated by machine learning methods.
Table 4.
Accuracies and MCC values obtained with various settings of k-NN, SVM, and ELM on the dataset consisting of the samples of PD patients whose UPDRS is below this threshold and 8 healthy subjects.
Table 5.
Best results obtained with k-NN, SVM and ELM with statistical significance tests.