Abstract: Anomaly detection algorithm based on robust multivariate probabilistic calibration model was proposed. This algorithm established normal status model of traffic flow matrix based on the latent variable probability model of multivariate t-distribution. The algorithm implemented network anomaly detection by comparing Mahalanobis distance between samples and normal status model. Theoretical analysis and experiments demonstrate its robustness and wide application. The algorithm is applicable when dealing with both data integrity and loss. It also has a stronger resistance over noise interference and lower sensitivity on model parameters, all of which indicate its performance stability.

Key words: anomaly detection; missing data; noise interference; probabilistic model; latent variable