Effective customer churn prediction (ECCP)‎ using parallel grey wolf optimizer

Other Title(s)

التوقع الفعال لإعراض الزبون عن الخدمة باستخدام محسن الذئب الرمادي الموازي

Dissertant

Dajani, Ahmad

Thesis advisor

Mafarjah, Majdi

University

Birzeit University

Faculty

Faculty of Engineering and Technology

Department

Department of Computer Science

University Country

Palestine (West Bank)

Degree

Master

Degree Date

2020

English Abstract

Metaheuristics algorithms gained the attention of many researchers in different optimization fields.

Feature Selection (FS) is combinatorial optimization problem where different metaheuristics algorithms were used to tackle it.

Grey Wolf Optimizer (GWO) is a recent population based metaheuristic algorithm that showed a good performance in tackling different optimization problems, including the FS problem.

As other global optimization algorithms, GWO suffers from a set of drawbacks (e.g.

stacking at the local optima and population diversity problem) that may degrade its performance.

In this thesis, a novel parallel GWO is proposed in the aim of maintaining a reasonable diversity of the population, in addition to helping the algorithm to escape the local optima.

Two parallel models were proposed; the first one called homogeneous GWO, where four copies of a GWO were employed on the same population.

While in the second approach, which is called heterogeneous GWO, four copies of the GWO, each one with a different updating strategy for the main parameter of the algorithm, were employed on the same population.

The proposed models were bench-marked on a set of well-known UCI datasets.

To assess the efficiency of the proposed algorithms, two experimental approaches were conducted.

The first experiment included comparing the proposed algorithms (independent, cooperative) with the original algorithm.

In this experiment, the independent homogeneous parallel version of the GWO showed a good performance when compared with sequential one, and the independent heterogeneous GWO outperformed both the sequential and homogeneous versions of the GWO.

Moreover, the cooperative heterogeneous algorithm outperformed all previous algorithms in terms of accuracy, however, it suffers from long execution time.

The second experiment included comparing the proposed algorithms with selected machine learning techniques (e.g CART) in term of accuracy.

Telecom company dataset from Kaggle data repository was used in this experiment to evaluate the proposed approach in order to predict the possible churner customers.

The results showed the superiority of the heterogeneous cooperative algorithm.

Main Subjects

Information Technology and Computer Science

Topics

No. of Pages

91

Table of Contents

Table of contents.

Abstract.

Abstract in Arabic.

Chapter One : Introduction.

Chapter Two : Literature review.

Chapter Three : Background.

Chapter Four : Parallel metaheuristics.

Chapter Five : The implementation of parallel BGWO.

Chapter Six : Experiments.

Chapter Seven : Conclusion and future work.

References.

American Psychological Association (APA)

Dajani, Ahmad. (2020). Effective customer churn prediction (ECCP) using parallel grey wolf optimizer. (Master's theses Theses and Dissertations Master). Birzeit University, Palestine (West Bank)
https://search.emarefa.net/detail/BIM-977602

Modern Language Association (MLA)

Dajani, Ahmad. Effective customer churn prediction (ECCP) using parallel grey wolf optimizer. (Master's theses Theses and Dissertations Master). Birzeit University. (2020).
https://search.emarefa.net/detail/BIM-977602

American Medical Association (AMA)

Dajani, Ahmad. (2020). Effective customer churn prediction (ECCP) using parallel grey wolf optimizer. (Master's theses Theses and Dissertations Master). Birzeit University, Palestine (West Bank)
https://search.emarefa.net/detail/BIM-977602

Language

English

Data Type

Arab Theses

Record ID

BIM-977602