Análise de performance de algoritmos estocásticos aplicados ao problema do caixeiro viajante

Abstract

Optimization algorithms are increasingly relevant tools in modern companies because they are capable of optimizing processes and resources, ensuring more efficient results and timely processing for decision making. Comparing these algorithms is a common process during their adoption studies. However, the use of complex methodologies can often lead to the choice of an imprecise algorithm, since its result may not reflect the reality of a company seeking to implement applications with limited resources. In view of this problem, the need arises to evaluate these algorithms from a new perspective. The main objective of this work is to propose a reflection on the way experiments in algorithms are conducted. The present study carried out experiments with optimization algorithms using computational resources similar to those found in most companies, comparing them with another work in which optimizations and tunings were used in these same algorithms. For the experiment, the traveling salesman problem was used, through 15 benchmarking divided into 3 categories, according to the size of each article. Finally, statistical metrics were obtained for the performance of each algorithm, which, when compared to the reference article, showed shorter execution times without compromising the accuracy of the results. Probabilistic algorithms are of great financial importance to companies that need to manage resources quickly, such as airports and shipyards. Therefore, the appropriate choice of parameters provides a more accurate view of reality.