Machine learning based classification of normal, slow and fast walking by extracting multimodal features from stride
Wajid Aziz1, Lal Hussain2,3, Ishtiaq Rasool Khan1
1Department of Computer & AI, College of Computer Science and Engineering (CCSE), University of Jeddah, P.O. Box 80327, Jeddah 21589, Saudi Arabia.
Mathematical Biosciences and Engineering : MBE
|February 2, 2021
Summary
Classifying walking speeds using machine learning accurately distinguishes normal, slow, and fast gaits. Ensemble classifiers like random forest (RF) and XGBoost show high performance in gait speed classification, aiding in understanding gait alterations.
Area of Science:
- Biomechanics
- Machine Learning
- Signal Processing
Background:
- Gait speed significantly influences biomechanical and spatiotemporal gait parameters across different age groups.
- Accurate classification of normal, slow, and fast walking is crucial for understanding gait alterations and their underlying mechanisms.
Purpose of the Study:
- To extract multimodal features from stride interval signals to classify normal, slow, and fast walking speeds.
- To evaluate the performance of various machine learning classifiers in distinguishing between different gait speeds.
Main Methods:
- Multimodal features, including time-domain and entropy-based complexity measures, were extracted from stride interval signals.
- Machine learning classifiers such as Classification and Regression Tree (CART), Support Vector Machine (SVM), Naïve Bayes (NB), Neural Network (NNET), and ensemble methods (Random Forest (RF), XGBoost, Averaged Neural Network (AVNET)) were employed.
- Performance metrics included accuracy, p-value, and Area Under the Receiver Operating Characteristic Curve (AUC).
Main Results:
- Random Forest (RF) and XGBoost achieved 100% accuracy, a p-value of 0.004, and an AUC of 1.00 for distinguishing slow from normal gait.
- Classifying fast and normal walking yielded up to 88% accuracy and an AUC of 0.94 with Naïve Bayes (NB).
- Distinguishing fast from slow gaits resulted in up to 88% accuracy and an AUC of 0.96 using XGBoost.
Conclusions:
- Machine learning models, particularly ensemble methods like RF and XGBoost, are highly effective in classifying different gait speeds.
- The findings provide a foundation for developing objective tools to assess gait abnormalities and support clinical evaluations.
