GBMOD: A granular-ball mean-shift outlier detector

Shitong Cheng, Xinyu Su, Baiyang Chen, Hongmei Chen, Dezhong Peng, Zhong Yuan

Abstract: Highlights•We construct coarse-granularity granular-balls from the data, effectively smoothing out noise through this representation.•We employ granular-balls to guide the mean-shift process and reduce the computational complexity.•We calculate outlier scores by measuring the distances between objects and the centers of their corresponding, shifted granular-balls.•Experimental results show that the proposed GBMOD algorithm has excellent accuracy, efficiency and robustness.