Authors:
Saman M. Almufti, Renas Rajab Asaad, Chen Jun, Yang Hao, Wang Fang, S. Sharan Jeev

Addresses:
Department of Information Technology, Technical College of Informatics-Akre, Akre University for Applied Sciences, Nineveh Governorate, Iraq.  Department of Engineering, Shanghai Tech Solutions, Shanghai, China. Department of Engineering, Dragon Peak Electronics, Sichuan, China. Department of Production, Golden Dragon Manufacturing, Zhejiang, China. Department of Cybersecurity, University of Texas at Dallas, Richardson, Texas, United States of America.

Abstract:

This paper applies the two-phase U-Turning Ant Colony Optimization (U-TACO) paradigm for the symmetric traveling salesperson problem to the binary 0–1 knapsack problem. The proposed formulation uses capacity-feasible constructive sampling, a normalized profit-to-weight heuristic, partial-subset U-turn reinforcement, elite and global-best pheromone deposits, and feasible add-and-swap improvement. Researchers used 30 diagnostic instance runs and 30 Pisinger instance runs (knapPI_1_100_1000_1) to evaluate eight parameter configurations, for a total of 480 algorithm runs. Each instance's configurations used the same seed set. Researchers verified the optimal diagnostic and benchmark results using exact dynamic programming and the Pisinger reference solution. Each design yielded earnings of 698 and 9,147, with 100% feasibility and no relative percentage variance. Analysis focused on iteration-to-optimal and runtime because the final objective values were equal. S4 had the lowest descriptive composite rank and reached the optimum in 1.500 iterations on the diagnostic instance, while S8 had the lowest mean runtime (0.037051 s) but took 4.933 iterations. Friedman tests demonstrated substantial configuration effects for both measures, and Holm-adjusted Wilcoxon tests identified the main pairwise differences. In the first iteration, all configurations met the Pisinger optimum; therefore, it was mostly an implementation and feasibility check. Direct numerical comparisons were limited to research utilizing the same instance, while a 30-method evidence matrix documented diverse techniques. S4 had the best convergence-cost balance and was kept as the working configuration. Scalability and superiority claims require evaluation across more Pisinger classes and item counts.

Keywords: 0–1 Knapsack Problem; Ant Colony Optimization; Binary Combinatorial Optimization; Parameter Sensitivity; Nonparametric Analysis; Pisinger Benchmark.

Received on: 17/09/2025, Revised on: 06/11/2025, Accepted on: 07/01/2026, Published on: 03/06/2026

DOI: 10.69888/FTSCIS.2026.000763

FMDB Transactions on Sustainable Critical Infrastructures, 2026 Vol. 1 No. 2, Pages: 73-88

• Views : 34
• Downloads : 6

Download PDF