Abstract: Anomaly detection is crucial in system maintenance.During operating process,normal operation data is easy to obtain,while anomal data usually take high cost to obtain.Therefore,one classifier could be utilized to solve the anmaly detection problem.Due to measurement uncertain,environment noise and storage problem etc.,uncertainness could be a characteristic of the montoring data.This paper utilized interval number to describe the uncertainess in the monitoring data,and raised an anomaly detection algorithm based on kernelized possibilistic 1-means clustering and 1-classifier for interval samples.The clustering center was considered both in the input space and the feature space.The interval width of the samples could be unbalanced,therefore,an interval splitting stragy was also proposed.Finally,illustrative numberic examples were given in utilizing artificial dataset and UCI machine learning repository.The effectiveness of the proposed algorithm is verified,and improvement is made by comparing with the existing SVM-OCC algorithms.

Key words: Interval samples,One cluster,One classifier,Interval splitting,Anomaly detection