A Study On Outlier Detection Of Seasonal And Trendy Spatial-temporal Data

With the in-depth development of mobile Internet and big data,countries,enterprises and individuals need to deal with more and more spatial-temporal data.Outlier detection has specific applications in many practical production and life,such as credit card fraud,industrial damage detection,and image detection.With the advent of the era of big data,it is becoming increasingly easy to obtain spatial-temporal data such as temperature data.The importance of detecting such outliers has also become increasingly prominent.Identifying outliers in spatial-temporal data has important implications for our lives.Spatial-temporal data has its own characteristics in the time dimension and space dimension.In the time dimension,the data tends to be characterized by trends,seasonality,periodicity,stability,autocorrelation,etc.In the spatial dimension,problems such as orderly disorder and limited boundaries are also to be distinguished.Because spatial-temporal data has many complicated situations in two dimensions,it is very challenging to link two dimensions effectively,but it is also necessary.Based on this,this paper proposes a spatial-temporal data outlier detection model based on local Gaussian process regression,effectively combining the two dimensions of space-time.First,the probability distribution of the predicted value is obtained through Gaussian process regression in consideration of seasonality and trend,and then the spatial domain identification method is used to determine whether the point is an abnormal value.Three sets of simulation data were constructed to fit three different spatial-temporal data conditions for testing.The results show that the outlier detection model based on local Gaussian process regression has high accuracy,from precision,recall,F1-Measures such as measure prove the validity of the model.