Outlier Detection And Applications Of Intervention Time Series Models

Due to errors,disturbances,or unusual events,time series generate values that are inconsistent with other observations.These values are called time series outliers.The existence of outliers has a significant impact on time series modeling and prediction,and outliers themselves also contain a lot of important information.Therefore,the mining and analysis of time series outliers is a significant task.The ARMAX model adds exogenous variables compared to the traditional ARMA model.It not only considers the variation of the sequence itself,but also considers the influence of other sequences.It is a more complete model.Based on the ARMAX model,the objective of this paper is to detect a single AO outlier and patches of AO outliers.In this paper,first,it introduces ARMAX model and model fitting steps.Then presents the use of standard Gibbs sampling to mine a single AO outlier and adaptive Gibbs sampling to mine patches of AO outliers.The effectiveness of these methods is verified by simulation.Experiment shows that the estimation of the outliers' magnitude and the finding of the outlier's position have achieved good results.At last,it selects the non-manufacturing business activity index and the whole society freight volume data,using these methods to detect three outliers in each set of data,and there are two consequent AO outliers in the whole social freight volume data.It analyzes and explains the reasons behind outliers in depth with the actual situation.