Loading...
Exploring the Use of Supervised Machine Learning Algorithms to Classify Simulated Balance Deficits
Sidener, Logan
Sidener, Logan
Citations
Altmetric:
Abstract
The purpose of this study was to determine the achievable accuracies of supervised machine learning algorithms for classifying simulated balance deficits and the number of participants needed to maximize those accuracies. The long-term goal is to create a classification system that can accurately detect the presence, severity, and progression of balance deficits in individuals who have a somatosensory deficiency. Postural sway data was collected from 27 healthy, young participants that had no significant history of balance disorders or musculoskeletal injury. Ground reaction forces and moments were collected while the individuals stood on force plates in two vision conditions, eyes open (EO) and eyes closed (EC), and five foam conditions, thicknesses of 0”, 1/8”, 1/4”, 1/2”, and 1”. These thicknesses were labeled as F0, F1, F2, F3, F4, respectively. The Center of Pressure (COP) was calculated and 30 linear and non-linear measures were calculated to characterize this time series. These measures were then used as the input features for 23 supervised machine learning algorithms with the goal of classifying the presence of the simulated balance deficit that differs in severity (i.e. classifying F0 vs F1, F0 vs F2, F0 vs F3, and F0 vs F4). Lastly, the number of participants needed to maximize the performance of each classification problem was determined. Maximum classification accuracies for each classification problem ranged from 56.79% to 80.86%, with the thicker foams being classified correctly at a higher rate. The most effective classification algorithm changed for each problem and data set, suggesting a wide range of algorithms should be explored in future work, and the eyes closed data set generally allowed for a higher classification accuracy than the eyes open data set. Finally, the number of participants needed to maximize the accuracy for each of the four classification problems was found to be 104, 147, 101, and 82. These results suggest that supervised learning algorithms can be used effectively to classify balance deficits and should be explored further with additional participants and individuals who suffer from a physiological balance deficit.
Description
Date
2018-01-01
Journal Title
Journal ISSN
Volume Title
Publisher
University of Kansas
Collections
Archive Status
This item contains archived web content.
Files
Sidener_ku_0099M_16138_DATA_1.pdf
Adobe PDF, 2.7 MB
- Embargoed until 2168-05-31
Research Projects
Organizational Units
Journal Issue
Keywords
Biomedical engineering, Biomechanics, Computer science, balance, deficit, learning, machine, postural, sway