Identification Of Outliers In Time Series Data

In regression analysis, data sets usually contain unusual observations that produces undesirable effects on least squares estimates, this unusual observations are refer to as outliers. Detecting these unusual observations prior data analysis is an important aspect of model building. However, many...

وصف كامل

محفوظ في:
التفاصيل البيبلوغرافية
المؤلف الرئيسي: Adewale Asiata Omotoyosi
التنسيق: أطروحة
اللغة:en_US
الموضوعات:
الوسوم: إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
الوصف
الملخص:In regression analysis, data sets usually contain unusual observations that produces undesirable effects on least squares estimates, this unusual observations are refer to as outliers. Detecting these unusual observations prior data analysis is an important aspect of model building. However, many regression diagnostics techniques have been introduced to detect these outliers. This research compares the performance of five regression diagnostics techniques based on Ordinary Least Square (OLS) estimators namely; standardized residuals, studentized residuals, Hadi's influence measure, Welsch Kuh distance and Cook's distance to detect and identify outliers. It is known that OLS is not robust in the presence of multiple outliers and high leverage points. Therefore, several robust regression models are used as alternative and its approach is more reliable and appropriate method for solving this problem. The robust regressions are M-estimation, Least Absolute Deviation (Ll), Least Median Square (LMS) and Least Trimmed Square (LTS). The comparisons are made via simulation studies and real data. This research also study the critical values of each techniques and our own critical values are computed for this research. Our results have shown that in some cases diagnostics based on OLS and some robust estimators give similar outcomes, they detect the same percentage of correct outlier detection. The results also shows that Least Trimmed Square is the best among all its counterparts followed by LMS, M estimator and L1 perform least.