Kernelized inner product-based discriminant analysis for interval data

This work presents an approach based on the kernelized discriminant analysis to classify symbolic interval data in nonlinearly separable problems. It is known that the use of kernels allows to map implicitly data into a high-dimensional space, called feature space; computing projections in this feature space results in a nonlinear separation in the input space that is equivalent to linear separating function in the feature space. In this work, the kernel matrix is obtained based on kernelized interval inner product. Experiments with synthetic interval data sets and an application with a Brazilian thermographic breast database demonstrate the usefulness of this approach.