MANAGING SOFTWARE PROJECT COSTS: TECHNIQUES AND BEST PRACTICES

Abstract

Cost management has been a long-time issue in software project development where cost overruns have been recorded to be more than 30-50 percent of the original budget as a result of poor estimation, scope creep as well as poor practices. The paper proposes a hybrid Function Point Analysis (FPA) and Machine Learning (ML) model that is able to predict costs of software projects classified into low, medium, and high categories to respond to deficiencies in other traditional methods, including COCOMO II, most of which depends on outdated and historical data. The methodology uses the existing FPA to perform the accurate functional sizing step and then apply the ML algorithms, like Random Forest and Support Vector Machines (SVM), to learn the required predictions. The accuracy on the model was 88.33 with a 0.89 (low), 0.86 (medium), and 0.94 (high) precision rates and a 0.91, 0.82, and 0.94 recall respectively. It has been verified using confusion matrices and training graphs that high-cost projects are robust. A comparative analysis shows better results than benchmarks: 6-13% accuracy improvements compared to COCOMO II, FPA-ANN (Jorgensen, 2014), and Agile Cost Models, due to the adaptability of ML to current complexities such as AI/ML projects and hybrid Agile-Waterfall environments. Combining predictive analytics and structured sizing, the method allows identifying early risks, allocating resources in the most efficient way possible, and making improved decisions, minimizing overruns and maximizing ROI. The additions to be made to it in the future will incorporate AI-specific features, deep learning, and the application of hybrid methodology to achieve greater scalability.