Statistical Modelling for Outlier Factors

Error in data is one of the facts that cause the parameter estimations to be subjective. If the erroneous case is proved statistically, then these cases are called outliers. Outliers are defined as the few observations or records which appear to be inconsistent with the rest of the group of the sample and more effective on prediction values. Isolated outliers may also have positive impact on the results of data analysis, data mining and estimated model. In this study, we are concerned with outliers in time series which have two special cases, innovational outlier (IO) and additive outlier (AO). The occurence of AO indicates that action is required, possibly to adjust the measuring instrument or mistake made by person in observation or record. However, if IO occurs, no adjustment of the measurement operation is required. Also in the study, a multi-factor ( 42 3 ) modelling was done in order to fit the effects of model in data analysis AR(1) coefficients, (0.5, 0.7, 0.9) outlier type (AO, IO), serie wideness (50, 100, 200, 500) and criterion value sensibility (% 99 (C=3.00), % 95 (C=3.50), % 90 (C=4.00)) factors statistically by making use of a simulation study. The results of the variance analysis on outlier factors were also emphasized.