Estimating null value using rough set theory and bees algorithm

العناوين الأخرى

تخمين القيمة المفقودة باستخدام نظرية المجموعة الصارمة و خوارزمية النحل

مقدم أطروحة جامعية

Shakir, Samir Adil

مشرف أطروحة جامعية

Duaymi, Mahdi Kazaz.
al-Ubaydi, Ahmad Tariq Sadiq

الجامعة

جامعة بغداد

الكلية

كلية العلوم

القسم الأكاديمي

قسم علوم الحاسبات

دولة الجامعة

العراق

الدرجة العلمية

ماجستير

تاريخ الدرجة العلمية

2012

الملخص الإنجليزي

Most of the real world databases are characterized by an unavoidable problem of incompleteness.

This thesis deals with null values problem that caused randomly in the databases by the user entries.

It presents a hybrid approach for solving the problem.

It hybridizes rough set theory with a swarm intelligence algorithm.

The proposed approach is a supervised learning model.

A large set of complete data called learning data is used to find the decision rule sets that have already been used in solving the incomplete data problem.

The intelligent swarm algorithm is used for feature selection which represents bees algorithm as heuristic search algorithm combined with rough set theory as evaluation function.

Furthermore another feature selection algorithm called Interactive Dichotomizer 3 (ID3) is presented that works as statistical algorithm instead of intelligent algorithm.

The proposed approach uses rough approximation sets in writing the rules; it writes rules with confidence equals to one from the lower approximation sets and rules with confidence less than one from the upper approximation sets.

The left hand side conditional attributes of the rules are selected using Bees’s evaluation function in the intelligent algorithm or ID3’s selection function in the statistical algorithm.

Reduct is the minimal subset of attributes that enables the same classification as the whole set of attributes.

So, reduct has been used as a stopping condition for the proposed approach.

In other words stopping the selection and writing rules operations are employed while the set of left hand side attribute represents a reduct.

In order to assess the system and show the performance of intelligent approach for null values estimation, a comparison between intelligent and statistical approaches is made in three parameters; rule set size, rule set II complexity and accuracy of estimation through working with rough set theory.

The results obtained from most code sets show that bees algorithm is better than ID3 in decreasing the number of extracted rules without affecting the accuracy and increasing the accuracy ratio of null values estimation, especially when the number of null values is increasing.

By using the best code set and in the case of one null value occurrence within each record of testing data, the proposed system has the ability of estimating 95.9% of null values as an approximate estimation and 91.6% of them as a true estimation.

On the other hand, when two null values occur within each record, the null values approximately estimated are 92.7% while 78.08% of them are absolutely true estimates.

The proposed system is also tested in the case of three null values occurrence within each record.

In this case the null values approximately estimated are 84.8% and 73.61% of them are absolutely true estimates.

التخصصات الرئيسية

تكنولوجيا المعلومات وعلم الحاسوب

عدد الصفحات

110

قائمة المحتويات

• APA
• MLA
• AMA

لغة النص

الإنجليزية

نوع البيانات

رسائل جامعية

رقم السجل

BIM-605415