A Comparative Study of Adaptive Crossover Operators for Genetic Algorithms to Resolve the Traveling Salesman Problem

Genetic algorithm includes some parameters that should be adjusting so that the algorithm can provide positive results. Crossover operators play very important role by constructing competitive Genetic Algorithms (GAs). In this paper, the basic conceptual features and specific characteristics of various crossover operators in the context of the Traveling Salesman Problem (TSP) are discussed. The results of experimental comparison of more than six different crossover operators for the TSP are presented. The experiment results show that OX operator enables to achieve a better solutions than other operators tested.

💡 Research Summary

The paper investigates how different crossover operators affect the performance of a Genetic Algorithm (GA) when solving the Traveling Salesman Problem (TSP). After a brief review of natural selection and the fundamentals of GAs, the authors describe the TSP as a combinatorial, NP‑hard problem whose search space consists of all permutations of n cities (size n!). They outline the standard GA pipeline—encoding, initial population generation, fitness evaluation, selection, crossover, mutation, and replacement—emphasizing that the choice of each component can dramatically influence solution quality and convergence speed.

For encoding, the study adopts the path (permutation) representation, where a chromosome is an ordered list of city indices. The initial population is created by three methods: pure random generation, mutation of a single random individual (N‑1 times), and a heuristic‑driven seed that starts from a city and repeatedly selects the nearest unvisited city, followed by mutation to diversify the rest of the population. Fitness is defined as the total tour length, calculated from a Euclidean distance matrix. Selection is performed with roulette‑wheel (fitness‑proportionate) selection, which the authors illustrate with a simple algorithmic pseudo‑code.

The core contribution is a comparative experimental study of more than six crossover operators adapted for permutation chromosomes. The operators examined include Ordered Crossover (OX), Partially Mapped Crossover (PMX), Cycle Crossover (CX), Edge Recombination Crossover (ERX), Position‑Based Crossover (PBX), and possibly others mentioned only briefly. Each operator’s mechanism is described: OX preserves a contiguous subsequence from one parent and fills the remaining positions in the order they appear in the second parent; PMX builds a mapping between two parent segments; CX creates cycles of gene positions; ERX attempts to preserve adjacency information; PBX selects random positions and copies them directly.

Experimental methodology: the authors fix all GA parameters (population size, number of generations, crossover probability Px, mutation probability Pm, and selection method) and vary only the crossover operator. They test on several benchmark TSP instances with city counts of 5, 10, 15, 20, and 25, running each configuration 30 times to obtain statistical averages. Performance metrics are (1) best tour length found, (2) average tour length across runs, and (3) number of generations required to reach convergence (or a predefined stopping criterion).

Results: Across all problem sizes, the Ordered Crossover (OX) consistently yields the lowest average tour length and reaches high‑quality solutions faster than the other operators. For medium‑sized instances (15–25 cities), OX’s advantage becomes more pronounced, showing both better final fitness and fewer generations to converge. PMX and CX occupy a middle ground, delivering respectable but not leading results. ERX and PBX display higher variability; in some runs they match OX, but on average they lag behind. The authors present tables and line graphs illustrating these trends, concluding that OX is the most effective crossover for permutation‑based TSP within the tested GA framework.

Critical appraisal: While the study provides clear empirical evidence favoring OX, several limitations temper the generality of the conclusions. First, the problem instances are relatively small and symmetric; real‑world routing problems often involve asymmetric distances, time windows, vehicle capacities, and stochastic elements, which could alter the relative performance of crossover operators. Second, the investigation isolates crossover as the sole variable; however, GA performance is known to be highly sensitive to the interaction among selection pressure, mutation rate, and adaptive parameter control. The paper does not explore whether alternative selection schemes (e.g., tournament, rank) or adaptive crossover probabilities might change the ranking of operators. Third, statistical rigor is limited; although multiple runs are performed, the authors do not report confidence intervals, hypothesis tests, or effect sizes, leaving the significance of observed differences somewhat ambiguous. Finally, the computational cost of each crossover (e.g., OX’s linear time versus PMX’s mapping overhead) is not quantified, which could be relevant for large‑scale applications.

Future work suggested by the authors includes extending the benchmark set to larger, more realistic TSP instances, incorporating additional constraints (time windows, vehicle capacities), testing hybrid crossover strategies, and applying adaptive mechanisms that adjust crossover and mutation probabilities on‑the‑fly based on population diversity metrics.

In summary, the paper contributes a systematic comparison of several permutation‑based crossover operators within a GA for the TSP, highlighting the superiority of Ordered Crossover under the experimental conditions. It serves as a useful reference for practitioners designing GA solvers for combinatorial routing problems, while also pointing out avenues for deeper, more comprehensive investigations.