Deep Neural Networks for Diagnosis of Parkinson's Disease Using Replicated Acoustic Features

Parkinson's disease is a neurodegenerative disease that affects a person's non-motor and motor functions causing tremors, stiffness, difficulty in moving, fatigue, anxiety, sleep disorders, etc. This condition happens because of the absence of dopamine-producing cells in the brain, and its symptoms worsen over time. This paper proposes a method for diagnosing Parkinson's disease using replicated acoustic features. This paper uses feature selection algorithms namely Fisher's, mRMR, Boruta, and ANOVA (Analysis of Variance) to find out the most discriminative features, and these features were fed into Random Forest (RF) and Deep neural networks (DNNs) to diagnose Parkinson's disease. 99.17% is the highest accuracy is achieved with the DNNs with Boruta and DNNs with ANOVA models. This is a significant improvement compared to previous studies. This result suggests that the replicated acoustic features can be used to diagnose Parkinson's disease.