Early detection of Parkinson disease using stacking ensemble method.

Area of Science:

• Neuroscience
• Medical Informatics
• Machine Learning

Background:

• Parkinson's disease (PD) is a progressive neurodegenerative disorder affecting dopamine-producing neurons in the substantia nigra.
• Accurate PD diagnosis is challenging due to data complexity, including imbalanced datasets and overlapping features between healthy and affected individuals.
• Existing single-classifier models often struggle with noisy data and can lead to overfitting, impacting diagnostic reliability.

Purpose of the Study:

• To develop an advanced ensemble-based model for improved Parkinson's disease diagnosis.
• To address the limitations of single-classifier models in handling noisy and imbalanced medical data.
• To enhance diagnostic accuracy and reduce bias and variance in Parkinson's disease detection.

Main Methods:

• Proposed an ensemble-based model named Ensembled Expert System for Diagnosis of Parkinson's Disease (EESDPD).
• Utilized a stacking ensemble technique to aggregate diverse predictive assumptions.
• Selected relevant features to enhance model performance.
• Compared EESDPD against several established machine learning models (Logistic Regression, SVM, Naïve Bayes, Random Forest, XGBoost, Decision Tree) and other PD diagnostic models (B-TDS-PD, B-TESM-PD).

Main Results:

• The EESDPD model demonstrated superior performance in classifying Parkinson's disease.
• Evaluated using key metrics: classification accuracy, precision, recall, and F1-score.
• The ensemble approach effectively managed data complexities that challenge single-model classifiers.

Conclusions:

• The proposed EESDPD model offers a robust and accurate solution for Parkinson's disease diagnosis.
• Ensemble methods, like the one proposed, are effective in overcoming the limitations of single classifiers for complex medical datasets.
• This approach holds promise for improving clinical decision-making in neurodegenerative disease diagnosis.