Abstract

High-level classification algorithms focus on the interactions betweeninstances. These produce a new form to evaluate and classify data. In thisprocess, the core is a complex network building methodology. The currentmethodologies use variations of kNN to produce these graphs. However, thesetechniques ignore some hidden patterns between attributes and requirenormalization to be accurate. In this paper, we propose a new methodology fornetwork building based on attribute-attribute interactions that do not requirenormalization. The current results show us that this approach improves theaccuracy of the high-level classification algorithm based on betweennesscentrality.

 

Quick Read (beta)