A Modified Value Difference Metric Kernel for Context-Dependent Classification Tasks

This paper focuses on the use of support vector machines on context-dependent classification tasks, like gene splice site prediction. For this type of problems, it has been shown that a context-based approach should be preferred over a transformation approach as this is a much more intuitive setting for working with distance functions. A weighted overlap metric and a modified value difference metric are described and subsequently used in several context-sensitive kernel functions. As a case-study gene splice site prediction is used and the experimental results show that the context-sensitive kernels making use of different weights and the modified value difference metric can always outperform their simple non-sensitive counterparts both in accuracy and in model complexity