Detection of Parkinson disease using multiclass machine learning approach

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Parkinson's Disease (PD) is a progressive neurological disorder impacting motor and cognitive functions, with early detection crucial for management.
• Current diagnostic methods may not capture subtle early-stage changes, necessitating advanced detection techniques.
• Voice signal alterations are recognized as potential early indicators of PD.

Purpose of the Study:

• To investigate the efficacy of Machine Learning (ML) and Deep Learning (DL) models in distinguishing individuals with Parkinson's Disease from healthy controls using voice recordings.
• To evaluate the performance of K-Nearest Neighbor (KNN) and Feed-forward Neural Network (FNN) models for PD detection.
• To optimize model performance through techniques like Synthetic Minority Over-sampling Technique (SMOTE), feature selection, and hyperparameter tuning.

Main Methods:

• Utilized a dataset of 195 voice recordings from 31 patients with Parkinson's Disease and healthy individuals, sourced from UCI.
• Applied ML/DL models including K-Nearest Neighbor (KNN) and Feed-forward Neural Network (FNN).
• Employed data preprocessing techniques such as SMOTE for class imbalance, feature selection, and RandomizedSearchCV for hyperparameter optimization.

Main Results:

• The Feed-forward Neural Network (FNN) model achieved superior performance with 99.11% accuracy, 98.78% recall, 99.96% precision, and a 99.23% F1-score.
• The Kernel Support Vector Machine (KSVM) model also demonstrated strong results, achieving 95.89% accuracy, 96.88% recall, 98.71% precision, and a 97.62% F1-score.
• Both FNN and KSVM models, trained on an 80-20 data split, proved effective in identifying Parkinson's Disease from voice signals.

Conclusions:

• ML and DL techniques, particularly FNN and KSVM, are highly effective for accurate Parkinson's Disease detection using voice analysis.
• Voice signal analysis presents a promising, non-invasive avenue for the early diagnosis of Parkinson's Disease.
• These findings highlight the potential of AI-driven approaches to significantly enhance early detection and intervention strategies for Parkinson's Disease.