Використання метрологічних підходів для верифікації експертних оцінок трудомісткості розробки пз

Abstract

This study presents a metrological framework for verifying expert software development effort estimates to reduce subjectivity in project planning. By treating requirement decomposition (feature lists) as an indirect measurement of software complexity, the authors implement an ensemble of hybrid neural networks acting as parallel measurement channels. Experimental validation showed an MMRE of 77.67% and RMSPE of 92.78%. Calculation of Type A standard uncertainty (56.11%) based on ensemble dispersion enables the use of statistical control and outlier detection to identify metrologically unreliable estimates. This approach transitions effort estimation from intuitive guessing to a verifiable statistical process with predictable confidence intervals.