Diverse Bagging Effort Estimation Model for Software Development Project

Creating successful projects is challenging and estimation of software development efforts is thus an important activity of the software engineering community. It enables project managers to organize and manage project quality, cost, resources, and timelines. However, standard techniques for estimating development effort struggle due to the increased complexity, dynamic requirements, multifaceted nature, non-linear relationship, and greater interdependencies of modern software. Various machine learning models have been created periodically to tackle the deficiencies of standard estimation techniques. Nevertheless, the deployment is limited due to inefficient model-constructing approaches and inconclusive results. By meticulously optimizing preprocessing and hyperparameter tuning steps, this research presents a Diverse Bagging Effort ESTimation (DBEEST) model for more reliable and accurate software development effort estimation. To accomplish this, six homogeneous ensembles through bagging were applied to the USP05-FT and SEERA datasets. Subsequently, the predictions of each homogeneous ensemble were combined through averaging to generate a more reliable and accurate prediction with improved robustness against inconsistencies and errors. The results demonstrate the DBEEST model outperformed all individual bagging ensembles and produced consistent results by delivering an overall average of low Mean Square Error, Root Mean Square Error, Mean Absolute Error, and Mean Magnitude Relative Error values and an overall average of high Coefficient of determination values across both diverse datasets. Moreover, the proposed model can improve efficiency in handling software development projects, resource optimization, facilitating informed decision-making, and on-time project completion.