Estimation of Distribution Algorithms Applied to the Next Release Problem

Abstract

The Next Release Problem (NRP) is a combinatorial optimization problem that aims to find a subset of software requirements to be delivered in the next software release, which maximize the satisfaction of a list of clients and minimize the effort required by developers to implement them. Previous studies have applied various metaheuristics and procedures, being many of them evolutionary algorithms. However, no Estimation of Distribution Algorithms (EDA) have been applied to the NRP. This subfamily of evolutionary algorithms, based on probability modelling, have been proved to obtain good results in problems where genetic algorithms struggle. In this paper we adapted two EDAs to tackle the multi-objective NRP, and compared them against widely used genetic algorithms. Results showed that EDA approaches have the potential to generate solutions of similar or even better quality than those of genetic algorithms in the most crowded areas of the Pareto front, while keeping a shorter execution time.