Sensor-based assessment using machine learning for predictive model of badminton skills
Badminton assessment is a process to evaluate the performance of players and it is very important for them to identify their strengths and weaknesses so as to improve their training effectiveness. Several conventional assessment methods, which are the lack of manpower, expertise and objective method...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2018
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/79265/1/ChewZhenShanMFKE2018.pdf http://eprints.utm.my/id/eprint/79265/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Badminton assessment is a process to evaluate the performance of players and it is very important for them to identify their strengths and weaknesses so as to improve their training effectiveness. Several conventional assessment methods, which are the lack of manpower, expertise and objective methods. Besides, standard parameters and assessment model using machine learning for badminton assessment are still at research level. The main objective of this research is to design and develop a novel and effective system for badminton assessment . In this thesis, a total of three assessment modules (Module 1: Badminton Serving Accuracy, Module 2: Badminton Shots Quality, Module 3: Player’s Agility) were developed to extract the required measurable parameters of players through their serves, hits and agility. A 9 degree of freedom wireless sensor, an APDM Opal sensor and a badminton feedback sensor, XiaoYu 2.0 were used in this study to collect kinematic parameters such as acceleration , power and rotational speed. All the three modules were tested with 3 strong and 6 normal players and there were totally 46 collected features. A total of 39 out of 46 features have been proved being significantly different using t-test method. The three feature selection methods were named Relief, Principal Component Analysis and Correlation Feature Selection and were used for feature extraction. Then, the acquired datasets were tested by seven machine learning models , namely Random Tree (RT), Random Forest, Artificial Neural Network, K Star, Multiple Linear Regression, Gaussian Process and Support Vector Machine. Total of 21 assessment models had been constructed. The results show that the RT model produces prediction accuracy of 90.84% and correlation value of r=0.86. |
|---|