Enhancing User Experience Via Calibration Minimization using ML Techniques
Electromyogram (EMG) signals are used to recognize gestures that could be used for prosthetic-based and hands-free human computer interaction. Minimizing calibration times for users while preserving the accuracy, is one of the main challenges facing the practicality, user acceptance and spread of up...
Saved in:
Published in | International journal of advanced computer science & applications Vol. 14; no. 7 |
---|---|
Main Authors | , |
Format | Journal Article |
Language | English |
Published |
West Yorkshire
Science and Information (SAI) Organization Limited
2023
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | Electromyogram (EMG) signals are used to recognize gestures that could be used for prosthetic-based and hands-free human computer interaction. Minimizing calibration times for users while preserving the accuracy, is one of the main challenges facing the practicality, user acceptance and spread of upper limb movements’ detection systems. This paper studies the effect of minimized user involvement, thus user calibration time and effort, on the user-independent system accuracy. It exploits time based features extracted from EMG signals. One-versus-all kernel based Support Vector Machine (SVM) and K Nearest Neighbors (KNN) are used for classification. The experiments are conducted using a dataset having five subjects performing six distinct movements. Two experiments performed, one with complete user dependence condition and the other with the partial dependence. The results show that the involvement of at least two samples, representing around 2% of sample space, increase the performance by 62.6% in case of SVM, achieving accuracy result equal to 89.6% on average; while the involvement of at least three samples, representing around 3% of sample space, cause the increase by 50.6% in case of KNN, achieving accuracy result equal to 78.2% on average. The results confirmed the great impact on system accuracy when involving only small number of user samples in the model-building process using traditional classification methods. |
---|---|
ISSN: | 2158-107X 2156-5570 |
DOI: | 10.14569/IJACSA.2023.0140750 |