Study Of Spatio-temporal Outlier Detection

Outliers detection problems originated in the field of statistics,there are a lot of mature theory,such as based on the statistics,clustering,density,and deviation,etc.The fields of outliers detection application expanded from statistical area to many other areas,such as finance,weather,etc.With the development of sensor network,global positioning system(GPS)and handheld mobile devices,data acquisition is more and more convenient.Many scholars study spatio-temporal data,including spatio-temporal pattern mining,spatio-temporal clustering,spatio-temporal outlier detection,spatio-temporal prediction and classification,etc.Scholars have made some achievements on the aspect of the traffic data,the river data.However,how to apply traditional outlier data identification technique to spatio-temporal data is still a problem.Outlier detection of spatio-temporal data needs to consider the correlation of time series,and the impact of the adjacent point data on space.In this paper,using the Moving Average method in the time dimension,and on the spatial dimension Inverse Distance Weighted method,they are used to predict the observed value and residual,thus the problem of outlier detection of spatio-temporal data is transformed into two-dimensional residual outlier detection problem.Nearest neighbor based outlier detection algorithm measures the degree of outlierness of a point,which are compared to its neighborhood.Simulation results demonstrate effectiveness of the proposed algorithm.Finally,the algorithm is applied to detect outliers of air quality monitoring data and achieves good recognition effect.