**6. Clinical interpretation**

The binary output from a classifier implies that a clinician will need further tests to gather strong evidence at the time of taking a diagnosing decision if the patient presents PD or not. Thus, a deeper quantitative analysis of the results must be carried out if the binary results from multiple voice-based tests are contradictory.

For this reason, the work from Solana-Lavalle et al. [19], provides an analysis of the most important features used to classify a subject as a PD patient or control. By using *Principal Component Analysis* (PCA), the features, with the highest contribution to the detection of PD, were obtained and analyzed. It was found that the features which explained better the diagnosis result, for the case of female subjects, are related to higher frequencies, such as the 32nd and 33rd TQWT coefficients. On the other hand, for the case of male subjects, it is found that features, with the highest contribution to PD detection, are related to lower frequencies such as the fifth TQWT coefficient. The mean and the standard deviation of the most important features were computed for the PD.

Patients and controls and a comparison (PD patients vs. control) is done by using box-plots. According to the box plots, it is shown that there is a clear separation between both groups in most cases. This analysis could help the physician during the interpretation of a binary result to understand how much affected voice is, and the likeness that a patient belongs to one group or the other.
