Advancing Computational Mathematics through AI-Driven Optimization, Machine Learning Algorithms, and Intelligent Decision-Making Models
Computer and Decision Making: An International Journal · 04.11.2026 · 3 (6S) · 3802-3811
DOI: 10.1922/comdem.v3i6s.19199 ↗Kurz gefasst
Fachartikel in der Zeitschrift Computer and Decision Making: An International Journal, online erschienen am 04.11.2026. Verfasst von Dr. Mamta und 5 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
The vehicle routing problem with time windows (VRPTW) is a feasible decision problem that must be made under coupled capacity, scheduling and fleet constraints. While machine learning can help local search engine operators, it's not clear how much more it'll add to low cost adaptive selection. In this study, an operator-selection control (OSC) with three different controls, and the machine-learning-guided search (MGS) framework, are evaluated on the 56 Solomon 100-customer VRPTW instances. Seven feasible neighborhood operators were coupled with a random-forest reward predictor, a uniform-random policy, a training-selected fixed policy, and an online upper-confidence-bound (UCB) policy. A total of 13,765 operator decision records were used for training, which were only obtained from the optimization trajectories. Four different spatial-demand configurations were isolated by using a leave-one-geographic-configuration-out design, resulting in 672 held-out search runs, under a common budget of 3.5 k feasibility evaluations per search run. All returned routes were independently feasibility checked. The UCB policy had the lowest average fleet size of 8.500 vehicles, while the learning-guided and random policies had a fleet size of 8.536 and the fixed policy, a fleet size of 8.613. Learning guided search outperformed the fixed policy in 102 out of 168 pairs of runs, and did not outperform UCB, and was more time consuming. The results show a practical approach towards combining mathematical optimization and machine learning, and highlight the importance of comparing learned decisions with economically viable adaptive options.
Zitieren
APA
Mamta, D., K.Rajkumar, A, S. C., Makandar, D., Parashar, N., & Chaduvula, R. J. (2026). Advancing Computational Mathematics through AI-Driven Optimization, Machine Learning Algorithms, and Intelligent Decision-Making Models. Computer and Decision Making: An International Journal, 3(6S), 3802-3811. https://doi.org/10.1922/comdem.v3i6s.19199
Vancouver
Mamta D, K.Rajkumar, A SC, Makandar D, Parashar N, Chaduvula RJ. Advancing Computational Mathematics through AI-Driven Optimization, Machine Learning Algorithms, and Intelligent Decision-Making Models. Computer and Decision Making: An International Journal. 2026;3(6S):3802-3811. doi:10.1922/comdem.v3i6s.19199
BibTeX
@article{Mamta2026,
author = {Mamta, Dr. and K.Rajkumar and A, Savariyammal Caroline and Makandar, Dr.Mujir and Parashar, Nishtha and Chaduvula, Rathna Jyothi},
title = {Advancing Computational Mathematics through AI-Driven Optimization, Machine Learning Algorithms, and Intelligent Decision-Making Models},
journal = {Computer and Decision Making: An International Journal},
year = {2026},
volume = {3},
number = {6S},
pages = {3802--3811},
doi = {10.1922/comdem.v3i6s.19199}
}
EUVIMED