Weighted Smooth Support Vector Machine (wSSVM) for Semi-supervised Learning.

• Main Authors: I-Ting Hung, 洪翊庭
• Other Authors: 黃文瀚
• Format: Others
• Language: en_US
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/94207068649106655838

碩士 === 國立中興大學 === 統計學研究所 === 103 === Despite a large amount and variety of data is now available on the internet, inevitably there exist difficulty in collecting data exhaustively due to a limited budget. Regarding a classification problem, we consider a semi-supervised learning model on a dataset where some subjects are not labeled but assume that the unlabeled proportion of each category is known. Taking this information about the unlabeled proportions into account, we propose a weighted smooth support vector machine (wSSVM) for the semi-supervised learning model. The strengths of the wSSVM method are to improve the accuracy of classification and ease of implement. To evaluate the performance of the proposed method, we conduct a simulation study from two perspectives of generating artificial data and resampling real data.