Software defect prediction based on adaptive supervised machine learning

العناوين الأخرى

تنبؤ خلل البرمجيات وفقا للتعلم الآلي الموجه

مقدم أطروحة جامعية

al-Rawashidah, Hanin Abd al-Sayyid

مشرف أطروحة جامعية

al-Abbadi, Muhammad Ali Husayn

أعضاء اللجنة

al-Kasasibah, Muhammad Sharari Zamil
Sulayman, Hamzah Sabah Iyal
al-Hammuri, Awni Mansur

الجامعة

جامعة مؤتة

الكلية

كلية تكنولوجيا المعلومات

القسم الأكاديمي

قسم الحاسوب

دولة الجامعة

الأردن

الدرجة العلمية

ماجستير

تاريخ الدرجة العلمية

2017

الملخص الإنجليزي

The source code is considered the basic code for building any system used in our life.

Most automated systems are mainly built on the source code written in one of the programming languages.

The rapid increase in the automated systems is equivalent with the need to write more source codes, so the efficiency of any system and its ability to carry out the tasks assigned to it depends on the absence of the errors from the source code.

Software performance is considered the most important factor in determining the quality of the product or the system.

Therefore, the developers of the systems have increased their interest in the evaluation of the source code and predict the defect before it is implemented.

This step is very important in saving time, money, effort and raw components of the system.

This thesis aims to highlight the systems that predict the occurrence of a defect in any source code written in any programming language.

These systems face two main problems: the time consumed to predict the defect and the efficiency of this prediction.

To overcome these challenges, we have used Supervised Machine Learning to classify whether the source code contains a defect or not, depending on a set of features extracted from the source code which stored in database.

Our analytical study was based on four different datasets (cm1, pc1, kc1 and kc2).

Each of them contains a variable number of attributes and instances.

Five models were proposed for analytical evaluation experiments.

The first three models were based on improving the performance by modifying the properties of the same classifier.

The other two models were concerned with improving the system performance by reducing the dimensionality of data, cleaning the data and eliminating the noise from it.

In the initial parameter model, the default setting of six classifiers which are SVM, ANN, NB, RF, DT and KNN was used.

التخصصات الرئيسية

تكنولوجيا المعلومات وعلم الحاسوب

عدد الصفحات

54

قائمة المحتويات

• APA
• MLA
• AMA

لغة النص

الإنجليزية

نوع البيانات

رسائل جامعية

رقم السجل

BIM-780520