Adaptive Machine Learning Approach for Parkinson’s Disease Diagnosis via Clustering-Based Speech Feature Selection

Abstract This study introduces a detailed approach aimed at improving the diagnostic precision of Parkinson’s disease by leveraging machine learning techniques applied to Parkinson’s Data. The dataset underwent thorough preprocessing, which included checks for missing values and the use of a Box-Cox transformation to normalize features. Following this step, the data was randomized and divided into training, validation, and test sets. Various feature extraction methods were investigated, such as Mutual Information, Recursive Feature Elimination, Lasso Regression, Clustering Feature Selection, and Autoencoders. These methods were paired with several regression models, including Ridge Regressor, Lasso Regressor, Linear Regressor, Decision Tree Regressor, k-Nearest Neighbors Regressor, Random Forest Regressor, Gradient Boosting Regressor, AdaBoost Regressor, and Extra Trees Regressor. The research found that combining Clustering Feature Selection with the Extra Trees Regressor yielded the highest accuracy, achieving an outstanding R 2 score of 99.9% on the Motor Test and 99.6% on the Total Test. These findings not only demonstrate the effectiveness of the proposed method but also establish a new benchmark in Parkinson’s disease diagnosis, significantly 1 surpassing existing state-of-the-art techniques. The results suggest that the integration of sophisticated feature selection and robust machine learning models can substantially enhance diagnostic accuracy in neurodegenerative diseases.