Abstract: Outlier detection and repairing in a data stream environment are discussed.An online detection method for outliers over data streams, called amnesia Kalman filtering (AKF), is proposed.In order to identify outliers, an improved Kalman filter is applied with the amnesia factor to forecast data-value in the future times tamp. A novel online adaptive repairing method for outliers over data streams, called adaptive interpolating wavelet (AdaptiveIW), is then proposed.The AdaptiveIW applies a method of variable-resolution interpolation, named the interpolation wavelet with adaptive resolution, to repair outliers, which determines interpolation resolution based on the number of continuous outliers.It adapts to different requested precision for out lier repairing over evolving data streams very well.Experimental results on actual power load data show that this method can provide precise instantaneous detection and accurate repairing for outliers over data streams.

Key words: adaptive, data streams, outlier detection and repairing, Kalman filtering, interpolating wavelet