Fast fixed granular-ball for attribute reduction in label noise environments and its application in medical diagnosis

lthough neighborhood rough set(NRS) based attribute reduction methods have achieved excellent performance in many scenarios, the efficiency and robustness of these methods have not attracted much attention. In this study, we propose a fast fixed granular-ball model (FFGB) for attribute reduction in label noise environments. In FFGB, we propose a fast neighborhood search mechanism to improve the efficiency of NRS. This fast mechanism reduces the neighborhood search range from the universe to a neighborhood and reduces the time complexity of the neighborhood calculation to much less than O(n^2) . Based on the fast mechanism, we propose FFGB model whose definitions are relaxed to be robust to against label noise. In addition, a FFGB attribute reduction algorithm is designed. Finally, we apply the FFGB attribute reduction to medical diagnosis. The experimental results indicate that FFGB is more efficient and robust than the comparison methods.