Manifold Learning and Application on Classification of Leukemia Cells Based on Raman Spectroscopy

Many machine learning technique have been employed for the classification of biological cells based on their Raman spectroscopy. Unfortunately, Raman spectroscopy data always has so many attributes that people who deal with them may often confront the problem of "curse of dimensionality". PCA is often used as a linear dimensionality reduction technique for preprocessing Raman data, which will not always get the satisfactory result, because Raman spectroscopy data are often distributed in nonlinear space. In this paper, we proposed a novel supervised nonlinear dimensionality reduction technique, KSISOMAP, which introduce kernel function and supervised learning to the ISOMAP, the experiments on Raman spectroscopy data and UCI data reveal that the KS-ISOMAP is a promising method and excellent in classification for Raman spectroscopy of leukemia cells.