Evaluating mean squared error as a fitness function in SOMA for software effort estimation: Insights from the NASA dataset

Abstract

Software effort estimation plays a pivotal role in software development. The Constructive Cost Model (COCOMO) is one of the most well-known algorithmic models for estimating software effort. However, the precision of these estimates is susceptible to input constants, potentially resulting in inaccuracies. To address this challenge, this research employs the Self-Organizing Migration Algorithm (SOMA), a metaheuristic algorithm, to optimize input constants of the basic COCOMO. This study uses the NASA18 dataset to evaluate the proposed experiment’s performance against the original COCOMO. Evaluation criteria such as MMRE, PRED(0.25), MAE, and MSE, with MSE serving as a fitness function, were employed to validate results. Comparative analysis indicates that optimized COCOMO estimates exhibit improved prediction accuracy. Building on these promising findings, future research will extend testing more deeply with other datasets and involve investigation of the intermediate COCOMO and COCOMO II.