Prediction of Parkinson’s disease using Ensemble Machine Learning classification from acoustic analysis

Parkinson’s disease (PD) is a neurodegenerative disorder, which upon progression affects the movements. Tremors associated with Parkinson’s disease are the major symptoms to look out for in such cases. This generally results in breakdown of the neurons producing dopamine. Qualitative speech starts to decline as the disease progresses and the variability in the vocal cord vibration (also known as fundamental frequency) starts to occur. People with PD have shown to produce greater variability in frequency as compared to normal people. In our paper, we are focused on comparison of the voice measurement features of patient dataset to understand whether a patient is suffering from PD or not using Machine Learning classifiers. We have implemented Decision Trees, Logistic Regression and K-nearest neighbors as base classifiers and have compared their performance with Ensemble learning classifiers Bagging, Random Forest and Boosting. We have compared the accuracy (%) of the classifiers and discussed which one of them is more accurate at predicting the outcome of the disease. We also found out the most relevant features associated with the classification and ranked them based on feature importance. Our main aim here is the classification of healthy individuals from people suffering from PD by detection of dysphonia (difficulty in speaking due to declining health conditions).