Distance-Based Outlier Detection for Distributed RFID Data Streams

Recently, RFID technologies have been widely used in many fields, such as real-time monitoring, object identification and item tracing, so it is very important to detect abnormal states of tagged objects in time. However, due to various environmental factors and unreliability of wireless communication technology, the data collected by the RFID readers are often noisy. According to the characteristics of distributed RFID data stream, such as huge-volume, variability, unreliability and distribution, the authors propose a distance-based local stream outlier detection algorithm (LSOD) and an approximate estimate-based global stream outlier detection algorithm (GSOD). LSOD need maintain a data stream structure, CSL, to identify the safe inlier. With the help of the characteristics of safe inliers, LSOD not only can reduce the memory space of stream data, but also can save the query time on stream objects. Under the distance-based outlier definition, it is a fact that global outlier on center node is a subset of the union of local outliers on every distributed node. Thus, GSOD uses a sample-based method to estimate approximately the global outliers for reducing the communication volume and calculate load of the centre node. Finally, several experiments have been accomplished, confirming that the proposed algorithms have the characteristics such as short running time, small memory space and high accuracy.