Hybrid Feature Selection Framework for the Parkinson Imbalanced Dataset Prediction Problem.

Area of Science:

• Neurology
• Data Science
• Biomedical Engineering

Background:

• Parkinson's disease (PD) detection is crucial for early intervention, but imbalanced datasets in medical research often lead to classification bias.
• Neurological disorders like PD affect motor function, hearing, and cognition, necessitating robust diagnostic tools.

Purpose of the Study:

• To develop a hybrid feature selection framework to overcome data imbalance issues in Parkinson's disease detection.
• To enhance the accuracy and reliability of classification models for neurological disease diagnosis.

Main Methods:

• A hybrid framework combining Synthetic Minority Over-sampling Technique (SMOTE), Recursive Feature Elimination (RFE), and Principal Component Analysis (PCA) was employed.
• Classification models including Support Vector Machine (SVM) and K-Nearest Neighbors (KNN) were trained and evaluated on PD acoustic datasets.

Main Results:

• The Support Vector Machine (SVM) model achieved a high accuracy of 98.2%, while the K-Nearest Neighbors (KNN) model demonstrated a specificity of 99%.
• The proposed method effectively addressed data bias, outperforming existing methods in Parkinson's disease prediction.

Conclusions:

• The developed system offers a significant advancement in early Parkinson's disease detection by mitigating data bias.
• This approach can aid health organizations in optimizing resource allocation and prioritizing patient care for neurological conditions.