الفهرس 
Detection and correction of outliers in time series data
AbstractOnly 14 pages are availabe for public view
 |  | from | 127 |  |  |
| | | from | 127 | | |
Abstract
There are many types of outliers that can occur in time series data, and they can be studied and detected using several tests and methods. In this study, various types of outliers were studied, which are innovational outliers, additive outliers, level shifts and variance changes. Various methods and tests were also reviewed and used to detect outliers in time series data. Two specific types of outliers were compared, which are innovational outliers and additive outliers in Autoregressive model. For this model, the likelihood function was studied and the maximum likelihood parameter was estimated for both cases: when the timepoint of the outlier is known and when this timepoint is unknown. It was concluded that, to detect either an innovational outlier or an additive outlier, time series data shall be treated as containing no outliers, then test the residuals of the parameters to detect the outlier, then model{u2019}s parameters shall be re-estimated after removing the effect of the outlier. The second part of this study focused on detecting innovational outliers and additive outliers in autoregressive models of order 1, which means AR (1) models, in case of the existence of a constant value. Two different sample sizes of sizes 50 and 100 were simulated using SAS 9.2 package, different constant values were used for AR models, outliers were detected in the simulated data