The Rough Granular Approach to Classifier Synthesis by Means of SVM
Jacek Szypulski, Piotr Artiemjew · Lecture notes in computer science · 2015
In this work we exploit the effects of applying methods for constructions of granular reflections of decision systems developed up to now in the framework of rough mereology, along with kernel methods for the building of classifiers. In this preliminary report we present results obtained with the SVM classification with use of the RBF kernel. The approximation metod we use is the optimized $$\varepsilon $$ concept dependent granulation. We experimentally verify the validity of this new approach with test data: Wisconsin Diagnostic Breast Cancer, Fertility Diagnosis, Parkinson Disease and the Prognostic Wisconsin Breast Cancer Database. The results are very promising as the obtained accuracy is not diminished but the size of the granular decision system is radically diminished.