Outlier Detection Based on Local Density of Vector Dot Product in Data Stream

摘要

Outlier detection in data stream is an increasingly important research in many fields. To deal with the data stream with the properties of high dimension, rapid arrival in order, high cost of storing all data in memory and so on, an outlier detection algorithm based on local density of vector dot product in data stream (LDVP-OD) is proposed. LDVP-OD uses the model based on sliding window and multiple validations to decrease the false alarm rate, which divides the data stream into uniform-sized blocks. Local density of vector dot product (LDVP) is described in order to precisely evaluate the outlierness of data in data stream. Furthermore, an outlier judgment criterion based on supreme slope is introduced, which can determine the exact outliers without requiring the number of outliers or other parameters beforehand. Comparison experiments with existing algorithms on synthetic and real datasets prove the high detection rate, good stability, strong adaptability of LDVP-OD.