Machine Learning-Based Detection of Parkinson's Disease Using Vertical Ground Reaction Force and Gait Analysis

This study investigates the application of machine learning techniques for analyzing Vertical Ground Reaction Force (VGRF) and synthesized Stride data in Parkinson's disease (PD) detection. The study uses data from 93 Parkinson's disease patients and 73 control subjects, focusing on features including mean, standard deviation, skewness, energy, and entropy derived from VGRF and stride signals. Four classification models—Decision Tree, Efficient Logistic Regression, Ensemble, and Efficient Linear SVM—are utilized to analyze the data, focusing on metrics such as accuracy, sensitivity, specificity, and area under the ROC curve. The results show that VGRF signals reach a high accuracy of up to 83.66%, with the Decision Tree model performing the best. On the other hand, stride signal analysis is less effective but demonstrates the potential of the Efficient Linear SVM model with 70.0% accuracy. Combining VGRF and stride signals improves diagnostic accuracy, with the Efficient Logistic Regression model achieving an accuracy of 81.70%. The Efficient Logistic Regression model combining VGRF and Stride Signals in this study demonstrates superior performance, achieving a balanced approach in identifying both positive (Parkinson's disease) and negative (healthy controls) cases with an AUC of 0.8562, thereby establishing it as a robust choice for Parkinson's disease detection compared to models utilizing VGRF or stride signals alone.