Vocal Neurology: Powered Voice Screening for Parkinson’s Disease

In this work, we introduce an ensemble learning-based method for early detection of Parkinson's disease based on biomedical voice measurements. Voice impairment is one of the prevalent early symptoms of PD, and hence voice-based biomarkers prove to be an efficient way to detect the disease at an early stage. The data used in this study has different acoustic features derived from voice recordings, e.g., jitter, shimmer, harmonic-to-noise ratio, and pitch-related variations, which assist in the differentiation of PD patients from normal subjects. The data is preprocesses through noise removal, feature normalization, and data augmentation methods to enhance model performance and generalizability. In order to construct a sound predictive model, we use several machine learning classifiers such as Random Forest Classifier, Gradient Boosting Classifier, Ada Boost Classifier, Bagging Classifier, Extra Trees Classifier, Voting Classifier, Stacking Classifier, Logistic Regression, Decision Tree Classifier, K- Neighbors Classifier, and Gaussian NB. Each model is tested on a split dataset with some of it dedicated to training and the other to testing so that the models can learn from varied patterns in the data and we will submit the new known instance to predict. The individual models are compared against primary performance metrics like accuracy, precision, recall, F1-score. The Top Performing Models are put in Ensemble Learning.