IBED: Combining IBEA and DE for optimal feature selection in software product line engineering

• Main Authors: XUE, Yinxing, ZHONG, Jinghui, TAN, Tian Huat, LIU, Yang, CAI, Wentong, CHEN, Manman, SUN, Jun
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 2016
• Subjects: Optimal feature selection; Indicator-based evolutionary algorithm (IBEA); Differential evolutionary algorithm (DE); Software product line engineering; Software Engineering
• Online Access: https://ink.library.smu.edu.sg/sis_research/4969; https://ink.library.smu.edu.sg/context/sis_research/article/5972/viewcontent/IBED.pdf
• Institution: Singapore Management University

Software configuration, which aims to customize the software for different users (e.g., Linux kernel configuration), is an important and complicated task. In software product line engineering (SPLE), feature oriented domain analysis is adopted and feature model is used to guide the configuration of new product variants. In SPLE, product configuration is an optimal feature selection problem, which needs to find a set of features that have no conflicts and meanwhile achieve multiple design objectives (e.g., minimizing cost and maximizing the number of features). In previous studies, several multi-objective evolutionary algorithms (MOEAs) were used for the optimal feature selection problem and indicator-based evolutionary algorithm (IBEA) was proven to be the best MOEA for this problem. However, IBEA still suffers from the issues of correctness and diversity of found solutions. In this paper, we propose a dual-population evolutionary algorithm, named IBED, to achieve both correctness and diversity of solutions. In IBED, two populations are individually evolved with two different types of evolutionary operators, i.e., IBEA operators and differential evolution (DE) operators. Furthermore, we propose two enhancement techniques for existing MOEAs, namely the feedback-directed mechanism to fast find the correct solutions (e.g., solutions that satisfy the feature model constraints) and the preprocessing method to reduce the search space. Our empirical results have shown that IBED with the enhancement techniques can outperform several state-of-the-art MOEAs on most case studies in terms of correctness and diversity of found solutions.