Swarm Intelligence for Software Effort Estimation: An Empirical Study

Effort estimation is one of the most complex tasks in the software industry. Overestimates or underestimates can lead to a low quality of products and contract losses. Various estimation models are used for software development effort estimation; the Constructive Cost Model (COCOMO) is the most widely used one. In the literature different swarm intelligence (SI) algorithms have been implemented to enhance various estimation models. The aim of this work is to study the effectiveness of ten SI-based algorithms to improve the results of five estimation models, namely, the basic COCOMO model, Sheta models (called Sheta model 1 and Sheta model 2) and Uysal models (called Uysal model 1 and Uysal model 2). These models are evaluated on the NASA-18 dataset. The efficacy of the refined estimation models is evaluated using both performance metrics and the number of best-estimated projects, offering a dual evaluation that considers both theoretical accuracy and practice. The results show that the performance of the ten SI algorithms differs across each estimation model, and that the most efficient effort estimations rely on the model-algorithm combination. Compared to earlier studies, our enhanced models yielded the most accurate outcomes, reflecting the strength of our empirical study.