An Ensemble-Based Framework to Estimate Software Project Effort

Effort estimation is essential for successful software project planning, budgeting, and risk identification. However, the techniques used to estimate effort are often inaccurate, outdated, and only consider technical factors while neglecting project management or stakeholder engagement. Expert estimation remains an important technique for leveraging human expertise in software estimation, but solely relying on this technique causes biased and subjective predictions. Machine learning (ML) techniques have shifted the direction of software project effort estimation towards computational intelligence. Nonetheless, there is a lack of deployment due to ambiguous outcomes and ineffective modelbuilding approaches. This study presents an ensemble-based framework that can estimate software project effort more accurately with the incorporation of domain knowledge and experiences. To achieve this, six homogeneous classifier ensembles will be constructed using six distinct classifiers on the proposed USP05-FT dataset. The collected expert estimations will be integrated into the proposed dataset as an additional feature in the form of numerical values such as expert-provided software project effort estimations (in person hours) that provide additional insight and knowledge. Subsequently, the predictions of all six homogeneous classifier ensembles will be combined through majority voting to obtain a more accurate and reliable prediction with increased robustness against errors and uncertainties. The performance of the proposed framework will be evaluated using Recall, F-measure, Precision, and Accuracy. It is expected that the proposed ensemble-based framework for software project effort estimation will lead to more efficient and effective software project management, an improvement in resource allocation, empowering informed decision-making, and timely project delivery