A comparative study on evolutionary multi-objective algorithms for next release problem

The next release problem (NRP) refers to implementing the next release of software in the software industry regarding the expected revenues; specifically, constraints like limited budgets indicate that the total cost corresponding to the next software release should be minimized. This paper uses and investigates the comparative performance of nineteen state-of-the-art evolutionary multi-objective algorithms, including NSGA-II, rNSGA-II, NSGA-III, MOEAD, EFRRR, tDEA, KnEA, MOMBIII, SPEA2, RVEA, NNIA, HypE, ANSGA-III, BiGE, GrEA, IDBEA, SPEAR, SPEA2SDE, and MOPSO, that can tackle this problem. The problem was designed to maximize customer satisfaction and minimize the total required cost. Three indicators, namely hyper-volume (HV), spread, and runtime, were examined to compare the algorithms. Two types of datasets, i.e., classic and realistic data, from small to large scale, were also examined to verify the applicability of the results. Overall, NSGA-II exhibited the best CPU run time in all test scales, and, also, the results show that the HV and spread values of 1st and 2nd best algorithms (NNIA and SPEAR), for which most HV values for NNIA are bigger than 0.708 and smaller than 1, while the HV values for SPEAR vary between 0.706 and 0.708. Finally, the conclusion and direction for future works are discussed. & COPY; 2023 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

Automated summary

This paper uses and investigates nineteen state-of-the-art evolutionary multi-objective algorithms to compare their performance in solving the next release problem in the software industry. The problem aims to maximize customer satisfaction and minimize total cost. Three indicators, including hyper-volume, spread, and runtime, are examined to evaluate the algorithms. NSGA-II shows the best CPU run time and the results suggest that NNIA and SPEAR perform well within the reasonable range of hyper-volume and spread values. The conclusion and future research directions are also discussed.