A practical algorithm to outlier detection and data cleaning for the time-dependent signal

Outliers are the observations which do not follow the regular pattern of system, and usually generated by the mechanical faults, measurement error, changes in system behavior. For the precise demand with the dynamic modeling, it is necessary to remove outliers from the measured raw data before constructing the model. Traditional statistical parameter estimation often fail to detect outliers in the functional profile. In this paper we propose a practical approach to outlier detection, which are based on the characteristics of time-dependent signal. The proposed algorithm included the change rate of signal calculation, statistical parameters estimation, signal recovering. A bioinformatic application demonstrates powerfulness of the proposed algorithm. The most advantage of the propose method is that it only clean (i.e., detects and recovers) outliers and preserves all other information in the observations.