Least-squares support vector machines modelization for time-resolved spectroscopy

By use of time-resolved spectroscopy it is possible to separate light scattering effects from chemical absorption effects in samples. In the study of propagation of short light pulses in turbid samples the reduced scattering coefficient and the absorption coefficient are usually obtained by fitting diffusion or Monte Carlo models to the measured data by use of numerical optimization techniques. In this study we propose a prediction model obtained with a semiparametric modeling technique: the least-squares support vector machines. The main advantage of this technique is that it uses theoretical time dispersion curves during the calibration step. Predictions can then be performed by use of data measured on different kinds of sample, such as apples.