Bülent Üstün

Visualization and Recovery of the (Bio)chemical Interesting Variables in Data Analysis with Support Vector Machine Classification

Support vector machines (SVMs) have become a popular technique in the chemometrics and bioinformatics field, and other fields, for the classification of complex data sets. Especially because SVMs are able to model nonlinear relationships, the usage of this technique has increased substantially. This modeling is obtained by mapping the data in a higher-dimensional feature space. The disadvantage of such a transformation is, however, that information about the contribution of the original variables in the classification is lost. In this paper we introduce an innovative method which can retrieve the information about the variables of complex data sets. We apply the proposed method to several benchmark data sets and a metabolomics data set to illustrate that we can determine the contribution of the original variables in SVM classifications. The corresponding visualization of the contribution of the variables can assist in a better understanding of the underlying chemical or biological process.

Optimisation of support vector regression parameters

Near infrared (NIR) spectroscopy is a very popular technique for quantitative and qualitative analyses of a wide range of materials due its non-invasive and informative characteristics. Until recently, the partial least squares (PLS) regression technique was mostly used for quantitative analysis of NIR spectroscopic data because of its simplicity of use, speed and relative good performance. However, PLS is limited in modelling data sets containing strong non-linear relationships. In the past few years, support vector regression (SVR), an alternative regression technique, has become a very good candidate for quantitative analyses of NIR spectroscopic data due to its high generalisation performance and its ability to model non-linear relationships as well. Unfortunately, the use of SVR is limited because of its set of parameters that need to be optimised by the user. Variation of these parameters affects the generalisation performance of the SVR model: therefore it is necessary to find the optimal parameter settings. The problem of optimal parameter selection is further complicated by the fact that the SVR model generalisation performance depends on all these parameters together (interaction of parameters). This leads to the conclusion that a separate optimisation of each parameter is not sufficient to find the optimal SVR model. It might be possible to find the optimal parameter settings by a grid search optimisation method, but this makes the use of SVR quite cumbersome and very time-consuming. Furthermore, it will not always lead to an accurate SVR model. For this reason, we introduced a new optimisation approach based on genetic algorithms (GAs) and the simplex search techniques, which yield an accurate SVR model (i.e. optimal parameter settings) in a short period of time.