Improving the classification accuracy of the weld defect by chaos-search-based feature selection
To improve the classification accuracy of the weld defect, we suggest a chaos-search-based feature selection method (CSFS). We regard the selection of the features as an optimisation problem in which the classification accuracy is taken as the object function. Firstly, n feature candidates
are arranged in a predefined order and then an n-bit binary number is constructed, in which each bit represents whether the corresponding feature is selected or not. Secondly, the binary number is converted to a chaotic variable, and then chaos search is employed to find the feature subset.
Thirdly, classifiers are used to evaluate the performance of the feature subset found. Three classifiers are employed and the performance of CSFS is compared with that of a genetic-algorithm-based feature selection method (GAFS). The experimental result shows that the selection of the features
can improve the accuracy of the classification of the weld defect and that the best feature subset is classifier dependent. The average classification accuracy of the weld defect when using the feature subset obtained by CSFS is 2.22 higher than that obtained by CAFS; meanwhile, the computation
time is cut by 7.99.
Keywords: Feature selection; chaos search; classification; defects
Document Type: Research Article
Affiliations: Qingming Shen and Jianmin Gao are with the State Key Laboratory for Manufacturing System Engineering, Xi'an Jiaotong University, Xi'an, 710049, China.
Publication date: 01 October 2010
- Official Journal of The British Institute of Non-Destructive Testing - includes original research and development papers, technical and scientific reviews and case studies in the fields of NDT and CM.
- Information for Authors
- Submit a Paper
- Subscribe to this Title
- Information for Advertisers
- Terms & Conditions
- Ingenta Connect is not responsible for the content or availability of external websites
- Access Key
- Free content
- Partial Free content
- New content
- Open access content
- Partial Open access content
- Subscribed content
- Partial Subscribed content
- Free trial content