Search In this Thesis
   Search In this Thesis  
العنوان
Intelligent outlier identification and categorization in dynamic big data systems /
الناشر
Huda Mohammed Touny ,
المؤلف
Huda Mohammed Touny
هيئة الاعداد
باحث / Huda Mohammed Touny
مشرف / Ibrahim Farag
مشرف / Ahmed Shawky Moussa
مشرف / Ali S. Hadi
تاريخ النشر
2021
عدد الصفحات
103 Leaves :
اللغة
الإنجليزية
الدرجة
ماجستير
التخصص
العلوم الاجتماعية (متفرقات)
تاريخ الإجازة
04/12/2021
مكان الإجازة
جامعة القاهرة - كلية الحاسبات و المعلومات - Computer Science
الفهرس
Only 14 pages are availabe for public view

from 114

from 114

Abstract

Outlier detection has been a critical task of various application domains and has been researched for a while. Outlier detection represents a challenge as it is difficult to accurately define and quantify the notion of outliers. Another challenge lies in the customization of outlier detection to the corresponding domain. Thus, many techniques have been introduced for outlier detection, yet they do suffer drawbacks such as labelling a datum that is close to the separating boundary between normal and outlying behaviour. Hence, depending on a crisp cut-off value to identify outliers is not linguistically meaningful or insightful for reliable decision-making. In this research, five methods of fuzzy treatment for the Blocked Adaptive Computationallyefficient Outlier Nominator (BACON) algorithm are proposed rather than a crisp cutoff threshold. The proposed solutions use Fuzzy Computing to capture the intrinsic uncertainty around the border between the main-stream data and outliers.The experimentations done in this research are mainly divided into two sets.The first set of experiments concerns about fuzzifying the output of the last iteration of BACON. The other set of experiments concerns about the fuzzification of each intermediate iteration of BACON.The aim of conducting the first set of experiments is to analyze the levels of uncertainty of the candidate outliers obtained by BACON and how this may affect the interpretation of outliers. The motive for the other set of experiments is to investigate the possibility of reducing the number of iterations of BACON while still having approximate fuzzy intermediate set of outliers that matches the final set declared by BACON. Four repository datasets have been used in the experimental part of this research. The datasets are different in their characteristics to validate the proposed solutions under various scenarios. Moreover, large and big synthetic datasets have been used in testing the scalability of the proposed approaches for Big Data. The overall experimentation has concluded that the proposed fuzzy treatments for BACON provide more meaningful interpretations to the final results than its crisp version and captured the uncertainty at the boundary between the inliers and outliers of the data