Citation

Haris, Mohammad (2025) Ensemble machine learning model with expert knowledge integration for software development effort estimation. Masters thesis, Multimedia University.

Full text not available from this repository.

Abstract

As the demand for software escalates, development companies have to build the software within the stipulated budget and timeframe while meeting the client’s expectations. This can be achieved by estimating the effort needed to execute a project, as it provides methods to determine the requisite manpower, resources, duration, and cost for developing software. However, traditional Software Development Effort Estimations (SDEE) techniques falter due to modern software’s complexity, dynamic requirements, multifaceted nature, non-linear relationship, and greater interdependencies. Expert estimation remains an important traditional technique for leveraging human expertise in estimation, but exclusive reliance on this technique results in biased and subjective SDEE. Machine learning (ML) techniques have redirected SDEE toward computational intelligence. Nonetheless, deployment is hampered by inaccurate results due to the limitations of single ML models and insufficient model-construction methodologies. Additionally, the existing research studies neglect human involvement and rely entirely on ML estimations, which may overlook contextual factors and nuances that experts could provide based on their experience and domain knowledge. By meticulously optimizing crucial modelconstructing steps, this research study proposes a novel Stacking ensemble ML model, designated as EXPERT-Integrated DiverSe-Bagging EnSembles Stacking for SoftwarE DEvelopment Effort Estimation (Xpert-S3E4), which can estimate software development effort more reliably and accurately. To accomplish this, six diverse Bagging ensembles are employed as base estimators and integrate expert estimations to complement the estimations generated by the Bagging ensembles. Subsequently, a non-linear Feed-Forward Deep Neural Network (FFDNN) is employed as a metamodel that utilizes these integrated estimations as input to generate the final SDEE. The Xpert-S3E4 model is trained and evaluated separately on two prominent PROMISE repository datasets, USP05-FT and SEERA, with its performance rigorously evaluated by employing MSE, RMSE, MAE, and R². The results demonstrate that integrating expert knowledge improved the model’s ability to generate more accurate and reliable SDEE. In comparison to the five existing SDEE models, the Xpert-S3E4 model outperformed them by achieving lower values of MSE, RMSE, and MAE, as well as higher values of R² across both datasets.

Item Type:	Thesis (Masters)
Additional Information:	Call No.: Q325.5 .M64 2025
Uncontrolled Keywords:	Machine learning
Divisions:	Faculty of Computing and Informatics (FCI)
Depositing User:	Ms Nurul Iqtiani Ahmad
Date Deposited:	19 Jan 2026 03:58
Last Modified:	19 Jan 2026 03:58
URII:	http://shdl.mmu.edu.my/id/eprint/15189

Downloads

Edit (login required)