1. UKM Learning and Research Repository
2. PHD Thesis / Tesis PHD
3. Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Title:	Modified salp swarm algorithms with back propagation neural network for software engineering prediction problems
Authors:	Sofian Ahmad Ali Kassaymeh (P86165)
Supervisor:	Salwani Abdullah, Prof. Dr.
Keywords:	Universiti Kebangsaan Malaysia -- Dissertations Dissertations, Academic -- Malaysia Neural network Software engineering prediction Neural networks (Computer science)
Issue Date:	10-Mar-2021
Description:	Prediction is a type of supervised machine learning technique. It is also a statistical measure used to define the relationship between one dependent variable and a series of other mutable independent variables. Using prediction techniques, data can be processed to discover trends and predict events in any problem dataset. The solving of a prediction problem involves finding the function that will give the best (closest) output to all inputs. Recently, a variety of metaheuristic methods have been developed for use in prediction. However, no single method is suitable for different types of data. Moreover, many of the proposed methods suffer from technical problems, such as high dependency on suitable parameter values for a given problem and the extent of initial population diversity, in addition, to unbalance between intensification and diversification. The main goal of this research is to develop an approach for high-accuracy prediction of software engineering prediction problems (SEPPs), i.e., Software Development Effort Prediction (SEP), Software Test Effort Prediction (STP), and Software Fault Prediction (SFP). To achieve this goal, three methods are proposed. Three different types of datasets are used to evaluate the proposed methods. These datasets vary in complexity and size, and all 36 of them are taken from the PROMISE and GitHub repositories. First, this research investigates the integration of the Salp Swarm Algorithm (SSA) with the Backpropagation Neural Network (BPNN) for application to SEPP. The aim of this integration referred to as (SSA-BPNN) is to find the optimal parameters of the BPNN by using the SSA, and thereby, improve the prediction accuracy. The results of the evaluation of the proposed method reveal that SSA-BPNN is able to outperform the comparable methods in almost all SEPP datasets in respect of most performance measures. Next, this research develops a modified version of the SSA for optimization problems. The modifications concern the initial population diversity, the parameter tuning strategy, and hybridization. First, diversification of the SSA population referred to as (𝑆𝑆𝐴𝐻𝑑) is addressed to control exploration. Second, a new version of the SSA referred to as (𝑆𝑆𝐴𝐺𝐴−𝑡𝑢𝑛𝑒r ) is proposed to enhance the tuning parameters of the SSA using a self-adaptive parameter setting, where a genetic algorithm (GA) is adopted to find the optimal parameters for the SSA at each generation. Third, a modified algorithm is proposed by hybridization of SSA with a local search algorithm referred to as (HSSA). Evaluation results using several benchmark functions that vary between a uni-modal and multi-modal shows that all modifications have a positive impact on SSA performance. Furthermore, this research investigates the performance of all modified algorithms on the same SEPP datasets, with the same aim of finding the optimal parameters of the BPNN, to improve BPNN prediction accuracy. In addition, all the modified algorithms are compared against several state-of-the-art methods from the literature, and the results indicate the superiority of the modified algorithms. Also, the experimental results reveal that the modified algorithms can outperform the compared methods in most SEPP datasets in respect of all performance measures. Finally, a statistical significance test (Wilcoxon statistical test) was conducted, and results prove significance for the proposed methods in most obtained result.,Ph.D
Pages:	314
Publisher:	UKM, Bangi
Appears in Collections:	Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.