Title: A group lasso-based method with its application to biomedical spectroscopic data
Authors: Ying Zhu - National Institute of Education, Nanyang Technological University (Singapore) [presenting]
Abstract: High dimensional spectroscopic data consist of many overlapping absorption bands sensitive to the physical and chemical states of compounds and thus show highly correlated structure due to the complex system of biomedical spectroscopic data. In certain settings, there may exist distinct groups among the variables, and it could be more informative to exploit these groups when performing classification of samples. A model based on group lasso was developed by encouraging correlated spectral features within a group to have a shared association with the response. The proposed model characteristic with grouping effect yields more interpretable results in an application to biomedical spectroscopic data. The informative spectral absorption bands selected for classification have provided great evidence regarding the bioactive constituents of the samples.