Evaluation of Gender’s Effect in Predicting Parkinson’s Disease from Voice Recordings: A Random Forest Approach

Overview

Parkinson’s Disease (PD) is the second most prevalent neurological disease in the world, affecting more than 10 million people. It is characterized by a progressive loss of motor control, causing symptoms like tremors and impaired balance and eventually rendering the patient paralyzed and completely bedridden. PD has no cure, but an early diagnosis can help slow down the progression and improve the patient’s quality of life. However, the diagnosis of PD is often subjective and inaccurate because its presentation varies widely between individuals of different demographics. One problem presented by this variation is that PD voice-based detection tools are often trained primarily with male voices, resulting in a lack of accuracy in diagnoses for women. This study focuses on predicting the progression of PD from voice recordings and evaluates the variation by gender using a novel Random Forest Algorithm (RFA). The algorithm utilized a multivariate dataset extracted from the UCI Machine Learning data repository that consisted of 5,875 voice recordings from 42 subjects. The RFA introduced in this study both improves the accuracy for PD detection and establishes that diagnostic algorithms can consistently perform well across gender (99% accuracy for females and 97% for males). In comparison, previous machine learning approaches when accounting for differences across gender achieved a highest accuracy of 82.14%. Additionally, this study identifies age and gender-based differences in the expression profiles of voice parameters that can be useful in future clinical applications.