Development of Intelligent Parkinson Disease Detection System Based on Machine Learning Techniques Using Speech Signal

Abstract

Parkinson's disease is a brain condition that induces difficulty walking, standing, concentrating, trembling, and weakness. Parkinson's symptoms typically begin slowly and increase with time. Whenever the condition develops, individuals can experience trouble walking and communicating to others. Old people mostly tend to suffer from this disease and the number is expected to increase in the future. Machine learning (ML) techniques could help in the medical field in processing and analyzing data that offer good solutions in this field in terms of high accuracy and less required time compared to conventional methods. In this study, we proposed an enhanced methodology based on utilizing SMOTE to balance the dataset, due to the available dataset is imbalanced. then adopted extra tree classifier with k-fold technique after we balanced the dataset with SMOTE. we have achieved the best accuracy with respect to the classification accuracy in the literature, the obtained accuracy of our proposed model was higher than the used approaches in the related works. The new model for classifying the Parkinson's disease-dataset with class-imbalance data distribution achieved an accuracy of 96.52% by using our proposed method. The result shown that the dataset is lacked of balancing and it proves that the balancing in the dataset is important specially in medical classification. The impact of Optimal function selection, either automated by PCA or manually carried out, is clearly still being studied, and plays an essential role in improving the performance of machine learning.

Harnessing Voice Analysis and Machine Learning for Early Diagnosis of Parkinson's Disease: A Comparative Study Across Three Datasets

“…The study employed three datasets: two sourced from the reputable UCI ML Repository (16,17) and one from gshare (18), selected for their demonstrated reliability (11,12,(19)(20)(21)(22)(23)(24). Selection was based on data consistency.…”

“…For the RF classi er, we explored the number of trees (100, 125, 150), max features (auto, sqrt), max depth (4, 5, 6), criterion (gini, entropy), min samples split (8,10,12), and min samples leaf (4, 6, 8). Finally, for the SVM, we tested parameters C (0.125, 0.25, 0.5, 0.75) and gamma (0.1, 0.3, 0.6, 0.9).…”

“…This underscores the imperative for robust, objective voice evaluation methods (1,2,(5)(6)(7)(8). Recent advances in ML, including the use of Support Vector Machines (SVM), Random Forests (RF), and Deep Learning, have shown high accuracy in detecting subtle vocal changes characteristic of early PD (4, 5,[8][9][10][11][12][13][14].…”

Harnessing Voice Analysis and Machine Learning for Early Diagnosis of Parkinson's Disease: A Comprehensive Study Across Diverse Datasets

“…The selection rationale was predicated on the datasets' comprehensive and varied voice data and a minimum number of subjects (n) of 40, which is pivotal for enhancing the validation of our machine learning model across diverse data sources. These datasets are recurrently employed in analogous studies, underscoring their pertinence and reliability (11,12,(14)(15)(16)(17)(18)(19)).…”

“…For instance, A. Tsanas et al (2011) leveraged support vector machines and random forests with dysphonia features, marking a remarkable 99% accuracy (10). Additionally, Yuan et al (2023) and Thanoun et al (2021) have demonstrated diagnostic accuracies up to 95% and 96.52%, respectively, using machine learning algorithms on speech signal data (11,12). The methodological approaches vary, with some studies employing Support Vector Machines, Deep Neural Networks, and ensemble learning methods, often coupled with feature extraction and balancing techniques to improve classi cation accuracy.…”