Author(s): Wang Shitong | F.L. Chung | Deng Zhaohong | L.I.N. Qing | H.U. Dewen
Journal: Biotechnology
ISSN 1682-296X
Volume: 4;
Issue: 3;
Start page: 163;
Date: 2005;
VIEW PDF
DOWNLOAD PDF 
Original page
Keywords: Bioingormatics differential capability control machine | feature extraction | cancer gene expression datasets | classification
ABSTRACT
In this study, based on DCCM (Differential Capability Control Machine), the new feature-extraction criterion NFEC is developed using the first-order differential information and a new feature-extraction algorithm DCCFE is accordingly proposed for binary classification problems. NFEC and DCCFE are then extended to their multi-classification versions, i.e., m_NFEC and m_DCCFE, respectively. Present experimental results demonstrate that the new feature extraction criteria and algorithms outperform or have comparable performance with the current methods for cancer gene expression datasets. Furthermore, since the new algorithms here admit more general first-order differential functions as the basis functions instead of kernel functions in SVM-based method, they perhaps have more potential applications in bioinformatics in the future.
Journal: Biotechnology
ISSN 1682-296X
Volume: 4;
Issue: 3;
Start page: 163;
Date: 2005;
VIEW PDF
DOWNLOAD PDF Original pageKeywords: Bioingormatics differential capability control machine | feature extraction | cancer gene expression datasets | classification
ABSTRACT
In this study, based on DCCM (Differential Capability Control Machine), the new feature-extraction criterion NFEC is developed using the first-order differential information and a new feature-extraction algorithm DCCFE is accordingly proposed for binary classification problems. NFEC and DCCFE are then extended to their multi-classification versions, i.e., m_NFEC and m_DCCFE, respectively. Present experimental results demonstrate that the new feature extraction criteria and algorithms outperform or have comparable performance with the current methods for cancer gene expression datasets. Furthermore, since the new algorithms here admit more general first-order differential functions as the basis functions instead of kernel functions in SVM-based method, they perhaps have more potential applications in bioinformatics in the future.
