DOI: 10.38007/978-1-80053-485-8

Enterprise software projects face interrelated risks in delivery, quality, organization, and maintenance. Consistently assessing these risks is difficult when evidence is incomplete, indicators are interpreted differently, and expert judgments must be translated into timely management decisions. This monograph develops an assessment framework that integrates scale-specific indicators, supervised predictive modeling, a web-based prototype, and decision-support functions. Its central modeling task is to approximate a continuous assessment index from ten risk indicators. The index summarizes assessed exposure; it is not a calibrated probability of project failure or a monetary estimate of expected loss. 

Two ten-Indicator profiles are tailored for both larger projects and small and medium-sized projects. Both use five ordered input codes and a common four-grade reporting policy. A shared feedforward architecture is trained using baseline gradient descent, Adam, particle swarm initialization followed by backpropagation, and Levenberg–Marquardt optimization; random forest regression provides a comparison from a different model family. The larger benchmark contains 5,000 records, including 1,200 enterprise-derived records and 3,800 synthetic variants, split into a 4,000/1,000 development/test partition. The smaller benchmark contains 100 enterprise-derived records, split 70/30. Recorded errors favor Levenberg–Marquardt training in both settings. These are descriptive benchmark comparisons whose external validity depends on label quality, record independence, and the separation of model selection from final testing. 

Download Full Text