**5. Classification results in male and female populations**

In the work by Tsanas et al. [9] it was claimed that the problem of PD detection would be more adequate if the problem were addressed by conducting separate analyses on male and female subjects. At that time (2012), such experiments were not possible due to the reduced size of publicly available datasets. However, an adequate number of recordings, for separate statistical studies, has been currently available since the introduction of the dataset by Sakar et al. [7].

In the first work by Solana-Lavalle et al. [8], datasets for PD detection on male and female subjects, were unbalanced, i.e., the number of PD patients is greater than the number of controls. Experiments, with balanced sets of PD patients and controls, were later conducted by Solana-Lavalle et al. [6] with interesting results. *Analysis of Voice and Magnetic Resonance Images to Assist Diagnosis of Parkinson's Disease… DOI: http://dx.doi.org/10.5772/intechopen.99973*


### **Table 1.**

*Voice-based Parkinson detection.*

It was found that detection performance is increased if balanced datasets are used to train and test classifiers.

A comparison of the different methods of voice-based PD detection, proposed by the research community, is shown in **Table 1**. It is observed that different datasets have been analyzed; however, the largest one is the dataset introduced by Sakar et al. [4]. The method that achieves the highest detection performance with the largest dataset is the one proposed by Solana-Lavalle et al. [6]. In addition, to reach the highest detection performance, this method is characterized by the lowest feature vector size.
