A Study on Penalized Support Vector Machines
- 주제(키워드) SVM , Classification , Variable selection , SCAD , LASSO
- 발행기관 고려대학교 대학원
- 지도교수 구자용
- 발행년도 2011
- 학위수여년월 2011. 2
- 학위구분 석사
- 학과 일반대학원 통계학과
- 세부전공 수리통계학 전공
- 원문페이지 28 p
- 실제URI http://www.dcollection.net/handler/korea/000000026019
- 본문언어 영어
- 제출원본 000045640580
초록/요약
In statistical modeling, classification and variable selection are fundamental problems. Although support vector machines have been effective tools in classification problems especially for high-dimensional data, they poorly performs when there are noisy variables. To solve this problem, diverse penalty functions have been suggested. This thesis mainly deals with representative penalized support vector machines, SCAD SVM and L1 SVM. Their performance is compared with simulation study. Diverse distributions of input variables were used in order to compare their performance. And then real data analysis was conducted. I finish this thesis as mentioning and suggesting what method would be appropriate for certain cases.
more목차
1 Introduction 1
2 Penalized SVMs 3
2.1 L1 SVM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 SCAD SVM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3 Experimental Results 8
3.1 Performance Measure . . . . . . . . . . . . . . . . . . . . . . . . 8
3.2 Simulation Study . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2.1 Normal mixture case . . . . . . . . . . . . . . . . . . . . 9
3.2.2 Inseparable multivariate normal case1 . . . . . . . . . . . 10
3.2.3 Inseparable multivariate normal case2 . . . . . . . . . . . 11
3.2.4 Separable multivariate normal case1 . . . . . . . . . . . . 12
3.2.5 Separable multivariate normal case2 . . . . . . . . . . . . 14
4 Data Analysis and Application 16
4.1 Wisconsin Breast Cancer Data . . . . . . . . . . . . . . . . . . . 16
4.2 Comparison of Performance for WDBC Data . . . . . . . . . . . 17
5 Conclusion 19
References 20
