An information-geometrical approach to kernel construction in SVM and its application in soft-sensor modeling

The method of soft-sensor modeling based on support vector machine is first analysed. Geometry of kernel function is studied from information geometry perspective in view of important influence of kernel's type on the performance of support vector machine. Then the kernel function is constructed in data-dependent way in order to improve the performance of support vector machine. Simulation results for both artificial data and real application show the effectiveness of the proposed method.