Purpose Parkinson’s disease (PD) is a progressive neurodegenerative disorder that primarily affects motor function, leading to symptoms such as tremors, rigidity, and bradykinesia. Early detection is crucial for timely intervention and improved disease management. Methods This study investigates the use of gait data and discrete wavelet transform (DWT) based feature extraction for PD detection. To classify extracted features, a feed-forward neural network (FFNN) was proposed alongside evaluating traditional machine learning algorithms, including support vector machines, random forests, decision trees, Naive Bayes, and XGBoost. Gait data was subjected to DWT to obtain time–frequency representations, and variance-based feature selection was employed to reduce dimensionality while preserving discriminative information. A comparative analysis was conducted to evaluate the performance of the proposed FFNN and ML models on statistical features derived from both raw gait data and DWT-transformed features, as well as their reduced feature sets. Results The proposed FFNN achieved test accuracies of 91% on raw data and 98% on DWT-transformed data. Under K-fold cross-validation, the FFNN attained the highest test accuracy (100%) in multiple folds. Additionally, approximation coefficients of high variance feature (with <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="521_2025_11621_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="49" /> </InlineMediaObject> <EquationSource Format="TEX">$\approx {96\%}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mo>≈</mo> <mrow> <mn>96</mn> <mo>%</mo> </mrow> </mrow> </math></EquationSource> </InlineEquation> data reduction from all the DWT features) achieved a test accuracy of 91% and reached 100% accuracy in one of the cross-validation folds. Conclusion These findings highlight the potential of variance-based feature selection and DWT-based feature extraction in enhancing automated PD detection using gait data.

Gait data classification for Parkinson’s disease detection using discrete wavelet transform features and a proposed neural network model

期刊论文
发表日期：

2025年9月12日

查看全文

Ayeesha Banu W,
N. Shilpa

摘要

Purpose

Parkinson’s disease (PD) is a progressive neurodegenerative disorder that primarily affects motor function, leading to symptoms such as tremors, rigidity, and bradykinesia. Early detection is crucial for timely intervention and improved disease management.

Methods

This study investigates the use of gait data and discrete wavelet transform (DWT) based feature extraction for PD detection. To classify extracted features, a feed-forward neural network (FFNN) was proposed alongside evaluating traditional machine learning algorithms, including support vector machines, random forests, decision trees, Naive Bayes, and XGBoost. Gait data was subjected to DWT to obtain time–frequency representations, and variance-based feature selection was employed to reduce dimensionality while preserving discriminative information. A comparative analysis was conducted to evaluate the performance of the proposed FFNN and ML models on statistical features derived from both raw gait data and DWT-transformed features, as well as their reduced feature sets.

Results

The proposed FFNN achieved test accuracies of 91% on raw data and 98% on DWT-transformed data. Under K-fold cross-validation, the FFNN attained the highest test accuracy (100%) in multiple folds. Additionally, approximation coefficients of high variance feature (with $\approx {96\%}$ $\approx 96 %$ data reduction from all the DWT features) achieved a test accuracy of 91% and reached 100% accuracy in one of the cross-validation folds.

Conclusion

These findings highlight the potential of variance-based feature selection and DWT-based feature extraction in enhancing automated PD detection using gait data.