MACHINE LEARNING APPROACHES FOR EARLY DIAGNOSIS OF PARKINSON'S DISEASE: A COMPARATIVE STUDY AND MODEL OPTIMIZATION

Abstract

Parkinson’s disease is the neurological illness that is comprises of both motor and non-motor symptoms which affects the population worldwide. Symptoms like Bradykinesia, postural instability, muscle stiffness, cognitive dysfunction and speech impairments are observed in the patients having PD. As early diagnosis is important for disease management but it is quite difficult to achieve when the PD symptoms are mild that causes a delay in clinical surveillance. This study offers a novel approach to improve diagnostic accuracy by using different ML algorithms on these acoustic features, extracted from Parkinson’s dataset would help in the early disease prediction. The ‘Clinical Parkinson’s dataset’ extracted from Kaggle, comprises of various vocal parameters like jitter, shimmer, nhr etc. which is used to predict Parkinson’s status by optimizing the UPDRS scores. Different Classification ML algorithms including Naïve Bayes, Logistic Regression, Random Forest, XG Boost, KNN and Deep Learning model i.e. ANN are implemented on the Parkinson’s dataset for PD detection and progression. Data preprocessing , feature selection and dataset splitting are crucial steps before the application of ML models. Splitting of dataset into 80/20 ratio for the training and testing, respectively, to check the model performance. This study reveals that the Deep Learning Model, ANN, shows the highest accuracy up to 97%, followed by XG Boost with 96%. This approach also helps in minimizing prediction errors. In addition to accuracy, there are certain other metric parameters like precision, recall and F-1 score which are used for model evaluation mainly in case of class imbalance. Incorporating voice-based data with effective ML models will facilitate the non-intrusive, and effective treatment of PD. This method holds the potential for remote precise and interpretable outcomes, resulting in early detection and enhanced patient outcomes