Kernel functions over orders of magnitude spaces by means of usual kernels. Application to measure financial credit risk

This paper lies within the domain of learning algorithms based on kernel functions, as in the case of Support Vector Machines. These algorithms provide good results in classification problems where the input data are not linearly separable. A kernel is constructed over the discrete structure of absolute orders of magnitude spaces. This kernel will be applied to measure firms' financial credit quality. A simple example that allows the kernel to be interpreted in terms of proximity of the patterns is presented.