Enhancing optimization algorithms with Kernel Density Estimation: A statistical learning strategy for smarter metaheuristics

Metaheuristic algorithms are commonly used for efficiently solving intractable optimization problems, given their ability to provide solutions while requiring relatively low computational efforts. This paper presents KDE-STOP, a novel strategy based on Kernel Density Estimation (KDE) to improve the efficiency of any target single-solution metaheuristic. In our framework, the algorithm analyzes objective function values attained during its initial iterations to estimate the distribution of encountered solutions. This distribution is then used to evaluate the probability of improving the current best solution, enabling the algorithm to reduce runtimes by terminating its execution as soon as it is unlikely to generate better solutions. The effectiveness of KDE-STOP is validated through experiments on a diverse set of metaheuristic and optimization problems, demonstrating its capability of improving the efficiency and effectiveness of a wide variety of target algorithms.