In Granular-ball Computing (GbC), the radius of a granular-ball is usually defined as the maximum or average distance from all enclosed objects to the center. However, both methods face challenges in building a high-quality family of granular-balls for enhanced classification performance. The former often results in overlaps between heterogeneous granular-balls, and the latter may fail to cover all objects. This paper presents an effective way to define the radius with adaptive granularity tuning and explores the subsequent application of the constructed granular-balls in classifications. Spec...
