Tuned Homogenous Ensemble Regressor Model for Early Diagnosis of Parkinson Disorder Based on Voice Features Modality

Abstract

Parkinson Disorder (PD) is a neurological disorder which is progressive and degenerative in nature. There are no specific tests pertaining to the diagnosis of PD. The symptoms at an early stage are mild. The early diagnosis of PD is really essential to delay the progression of the disorder. Speech disorder namely dysphonia is experienced by approximately 90% of PD patients. The incorporation of Artificial Intelligence (AI) techniques integrated with non-invasive capture of speech data from patients in diagnosis system aids to provide a robust, reliable and accurate estimation of Unified Parkinson Disease Rating Scale (UPDRS) score which ease the decision-making process effective for healthcare professionals. The proposed system incorporates a novel tuned Homogenous Ensemble Regressor wherein the hyperparameters are chosen and tuned using various experiments. Tuned Extreme Gradient (XgBoost) Regressor and Tuned Random Forest (RF) Regressor are the two homogenous regressor model. The proposed system is compared with the Tuned Linear Regression (LR) model which is the single Regressor model. The proposed system is evaluated using the large database of voice features samples of 42 PD patients. The Mean Absolute Error (MAE) and Mean Squared Error (MAE) values are minimal for the proposed system and it shows that the errors of the proposed system are lower than the single classifier errors and existing similar system.