A Comparison of Machine Learning Algorithms for Parkinson’s Disease Detection

Tóm tắt

Parkinson’s disease is a progressive condition that impacts the neurological system and the areas of the body that are under the control of the nerves. Detecting and diagnosing PD early and accurately is essential for effective treatment and care. Biomedical speech analysis has emerged as a promising noninvasive approach for detecting and monitoring PD. We collected a data set from the University of California at Irvine (UCI) machine learning repository. The study investigates the utilization of a dataset to assess the suitability of machine learning (ML) algorithms for diagnosing Parkinson’s disease. The research employs classification algorithms coupled with feature selection based on Pearson correlation. The data set is comprised of 195 voice recordings obtained from 31 patients during their examinations. Among them, 23 individuals in the study were diagnosed with Parkinson’s disease. The age at the time of diagnosis ranged from 46 to 85 years. To handle this data, three methodologies are employed: a train/test split, k-fold cross-validation, and stratified k-fold cross-validation to ensure robust evaluation. Additionally, we use the F1 score, accuracy, and balanced accuracy measures to evaluate our models. The highly correlated selection data combined with data processing methods helps achieve positive results for all machine learning algorithms. Especially, Random Forest consistently outperformed other methods in our project. Random Forest with an accuracy of 95.42%, balanced accuracy of 93.98%, and an f1-score of 0.98. This demonstrates the capability of algorithms to diagnose Parkinson’s disease effectively, even when confronted with imbalanced data. In our future research, we aim to integrate additional data on Parkinson’s Disease in order to conduct a comprehensive evaluation of this condition.