Abstract: The measured profile monitoring data of 48 wheels of a metro train is analyzed in batches. Taking the change rate of the flange wear area and the half change rate of the rolling circle radius difference at the flange root as analysis indexes, the boxplot algorithm and the improved isolated forest algorithm are used to detect the wear anomaly respectively. The boxplot algorithm has good anti-interference ability and can obtain objective statistical results of independent detection of each index. While the operation efficiency of the improved isolated forest algorithm is elevated. By comparison, the wear anomaly detection results obtained by the two algorithms are consistent, verifying the feasibility of both.