Finding boundary subjects for medical decision support with support vector machines

Support vector machines are learning machines designed to automatically deal with the accuracy/generalisation trade-off, by minimizing an upper bound on the generalisation error provided by VC theory. That makes them very attractive for applications in different domains, especially in the field of medical diagnoses. In the practice however there are still few tuneable parameters, which need to be set to accomplish best accuracy/generalisation trade-off. There are also some important design choices to select appropriate kernel, which transforms non-liner separable problems into high dimensional possibly linear separable problems. In this paper the influence of kernels and kernel parameters on classification accuracy is presented We also focus on the representation of knowledge extracted from support vector machine to make it usable for medical decision support.