Related Experiment Video
Updated: Aug 22, 2025
08:05
Design and Analysis for Fall Detection System Simplification
Published on: April 6, 2020
10.8K
Detection of Fall Risk in Multiple Sclerosis by Gait Analysis-An Innovative Approach Using Feature Selection Ensemble
Paula Schumann1, Maria Scholz2, Katrin Trentzsch2
1Institute of Biomedical Engineering, TU Dresden, Fetscherstr. 29, 01307 Dresden, Germany.
Brain Sciences
|November 11, 2022
Summary
Patient-reported walking questionnaires, like MSWS-12 and EMIQ, are most effective for predicting falls in people with Multiple Sclerosis (pwMS). Machine learning models, particularly Gaussian Naive Bayes, combined with feature selection, improve fall detection accuracy.
Area of Science:
- Neurology
- Rehabilitation Medicine
- Biomedical Engineering
Background:
- Walking impairment is a primary cause of falls in people with Multiple Sclerosis (pwMS).
- Gait analysis and fall detection are crucial for managing MS, with various clinical methods available.
- Combining gait analysis methods with machine learning for fall detection in pwMS remains underexplored.
Purpose of the Study:
- To identify the most effective methods for fall risk assessment in pwMS using machine learning.
- To determine the most significant features for fall detection in this population.
- To evaluate a novel feature selection ensemble (FS-Ensemble) for optimizing gait data analysis.
Main Methods:
- Analysis of eleven gait datasets using machine learning algorithms.
- Implementation of a feature selection ensemble (FS-Ensemble) incorporating Chi-square, information gain, Minimum Redundancy Maximum Relevance, and RelieF methods.
- Evaluation of four classification models: Gaussian Naive Bayes, Decision Tree, k-Nearest Neighbor, and Support Vector Machine.
- Utilized patient-reported outcome measures including the 12-item Multiple Sclerosis Walking Scale (MSWS-12) and Early Mobility Impairment Questionnaire (EMIQ).
Main Results:
- Patient-reported outcome measures (MSWS-12 and EMIQ) demonstrated the highest performance in fall detection, achieving an F1 score of 0.54.
- A combination of features from MSWS-12 and EMIQ, detailing mobility abilities and subjective walking experiences, yielded a recall of 75%.
- Gaussian Naive Bayes proved to be the most effective classification model across most datasets.
- The FS-Ensemble effectively improved classification performance and reduced feature set size.
Conclusions:
- Patient-reported walking questionnaires are highly relevant for fall risk assessment in pwMS.
- Machine learning, particularly Gaussian Naive Bayes coupled with FS-Ensemble, offers a promising approach for fall detection in MS.
- Further research should explore additional risk factors like fear of falling to enhance predictive models.
