Automated Diagnosis of Parkinson’s Disease using Speech Signals with Machine Learning

Parkinson's Disease ranks as the second most common chronic neurodegenerative condition that affects the CNS by killing the cells containing dopamine and its receptors. Dopamine is responsible for coordination and controls muscle activity, hence, individuals inflicted with Parkinson's Disease do unintended or involuntary movements due to lack of coordination. Non-motor symptoms of Parkinson’s Disease, also known as, dopamine-non-responsive symptoms encompass issues such as sleeping difficulties, constipation, drooling, swallowing and speech impairments. Notably, 90% of the diseased patients suffer from speech impairments, making it a viable sign to look at while diagnosis. Analyzing the acoustic measurements of diseased individuals can aid in the early diagnosis of Parkinson’s Disease, enhancing the efficacy of treatment. This study focuses on predicting Parkinson’s Disease based on the vocal analysis of individuals via Machine Learning based approach. This prediction is done by taking into account the various metrics of speech like frequency, amplitude, pitch, intensity and tonality that undergo alterations due to Parkinson’s Disease. Speech-based data from 31 subjects out of which 23 are diseased and 8 are healthy individuals is taken to create various data points for testing and validation. A comparative evaluation of various machine learning models is done for prediction and an ML-based methodology to diagnose Parkinson’s Disease in an individual with an accuracy of 96.15% is proposed solely on the basis of their voice structure and tonality.