Outlier Detection of Single Sensor based on Sliding Windows
Yin Long · Computer engineering & Software · 2014
Data quality is a major challenge of the web of things(Wo T),data quality can be improved and the underlying information can be mined by detecting outlier in Wo T data.The data spatial correlation is serious shortage in some small scale scenarios,such as smart home,it can only use the time correlation for single sensor data outlier detection.In this paper,a detection algorithm based on the distance outlier for sliding windows was given,the time complexity of the algorithm was reduced by only handling the new input instance and leaving instance,moreover,by just storing the k neighbors of the instance,the space complexity was reduced.Besides,based on the definition of local outlier and global outlier in sliding windows,this paper designed the process of outlier detection.The algorithm was simulated by using the real data from the smart home demo scenario,the detection rate(DR) and false alarm rate(FR) were the detection index of the algorithm to analyze the parameters affect of the detection results.The simulation results show that the algorithm can reach better detection results,local outlier detection can achieve higher DR,for global outlier to ensure low FR.