×

A novel support vector machine tree learning model. (Chinese. English summary) Zbl 1174.68528

Summary: The paper addresses a problem of redundant feature information contained in internal node of Confusion-crossed Support Vector Machine Tree (CSVMT) for classification in high dimensional feature space. A supervised locally linear embedding based CSVMT is proposed to involve the information between different features and correlation between data points in classification phase. Since local decision in each internal node requires different feature information, the proposed approach performs supervised locally linear embedding learning in each internal node on current assigned subset. The performance of the model is experimentally demonstrated with optical recognition of handwritten digits. The comparison results illuminate that the proposed learning model can achieve competitive recognition accuracy with more condensed structure than other compared models.

MSC:

68T05 Learning and adaptive systems in artificial intelligence
PDFBibTeX XMLCite