Parkinson's disease resting tremor severity classification using machine learning with resampling techniques.

Area of Science:

• Neurology
• Machine Learning
• Data Science

Background:

• Resting tremor, often caused by idiopathic Parkinson's disease (PD), is typically assessed using subjective methods like the Unified Parkinson's disease rating scale (UPDRS).
• Analyzing imbalanced datasets in PD severity classification presents a significant challenge for standard machine learning algorithms.
• Traditional PD evaluations are limited by their subjective nature and short observation periods of daily motor activities.

Purpose of the Study:

• To investigate the effectiveness of resampling techniques in improving the classification accuracy of imbalanced datasets for Parkinson's disease severity analysis.
• To compare the performance of under-sampling, over-sampling, and hybrid sampling methods when combined with various classifiers for PD data.

Main Methods:

• Employed resampling techniques, including under-sampling, over-sampling (random sampling), and hybrid combinations, to address class imbalance in PD patient data.
• Integrated these resampling methods with established classification algorithms: XGBoost, decision tree, and K-nearest neighbors.
• Evaluated classifier performance based on accuracy metrics.

Main Results:

• The over-sampling method demonstrated superior performance compared to under-sampling and hybrid techniques.
• Random over-sampling achieved 99% accuracy with the XGBoost classifier and 98% accuracy with the decision tree classifier.
• Observed that the effectiveness of resampling methods varied depending on the specific classifier used.

Conclusions:

• Over-sampling, particularly random sampling, is a highly effective strategy for enhancing the classification accuracy of imbalanced datasets in Parkinson's disease severity analysis.
• XGBoost and decision tree classifiers, when combined with appropriate over-sampling techniques, show significant potential for accurate PD severity assessment.
• The choice of resampling method and classifier is crucial for optimizing performance in PD data analysis.