Research Article
SVM Time Series Classification of Selected Gait Abnormalities
@INPROCEEDINGS{10.1007/978-3-030-95593-9_16, author={Jakob Rostovski and Andrei Krivošei and Alar Kuusik and Ulvi Ahmadov and Muhammad Mahtab Alam}, title={SVM Time Series Classification of Selected Gait Abnormalities}, proceedings={Body Area Networks. Smart IoT and Big Data for Intelligent Health Management. 16th EAI International Conference, BODYNETS 2021, Virtual Event, October 25-26, 2021, Proceedings}, proceedings_a={BODYNETS}, year={2022}, month={2}, keywords={Gait analysis Machine learning SVM Wearable sensors Medical applications}, doi={10.1007/978-3-030-95593-9_16} }- Jakob Rostovski
Andrei Krivošei
Alar Kuusik
Ulvi Ahmadov
Muhammad Mahtab Alam
Year: 2022
SVM Time Series Classification of Selected Gait Abnormalities
BODYNETS
Springer
DOI: 10.1007/978-3-030-95593-9_16
Abstract
Gait analysis is widely used for human disability level assessment, physiotherapeutic and medical treatment efficiency analysis. Wearable motion sensors are most widely used gait observation devices today. Automated detection of gait abnormalities, namely incorrect step patterns, would simplify the long term gait assessment and enable usage of corrective measures as passive and active physiotherapeutic assistive devices. Automatic detection of gait abnormalities with wearable devices is a complex task. Support Vector Machines (SVM) driven machine learning methods are quite widely used for motion signals classification. However, it is unknown how well actual implementations work for specific gait deviations of partially disabled people. In this work we evaluate how well SVM method works for detecting specific incorrect step patterns characteristics for the most frequent neuromuscular impairments. F1 score from 66% to 100% were achieved, depending on the gait type. Gait pattern deviations were simulated by the healthy volunteers. Angular speed motion data as an input to SVM was collected with a single Shimmer S3 wearable sensor.
