<![CDATA[The study addressed adaptive software-effort estimation under multifactor uncertainty by formalizing and evaluating a neuro-fuzzy project management model that integrated interval type-2 fuzzification, local weighted regression, Energy-Efficient BAT parameter optimization, and sequential online updating. Empirical verification used four public software-effort datasets: NASA93, Desharnais, Albrecht, and Kemerer. Dataset-specific predictors were processed independently, while software development effort was used as the target. All preprocessing, fuzzy partitioning, and parameter tuning were fitted only within training folds. Repeated five-fold cross-validation with ten repetitions and predict-then-update online adaptation were applied to prevent target leakage. The corrected model achieved a mean magnitude of relative error of 0.021±0.006, root mean square error of 0.011±0.003, prediction at 25% of 0.989±0.012, correlation of 0.969±0.010, and prediction index of 0.918±0.015. The results showed a balanced accuracy-adaptability trade-off relative to the selected baselines, while external enterprise validation remained necessary]]>
Copyrights © 2026
