Optimizing termination decision for meta-heuristic search techniques that converge to a static objective-value distribution

This paper proposes a new technique for assisting search technique optimizers (most evolutionary, swarm, and bio-mimicry algorithms) to get an informed decision about terminating the heuristic search process. Current termination/stopping criteria are based on pre-determined thresholds that cannot guarantee the quality of the achieved solution or its proximity to the optimum. So, deciding when to stop is more an art than a science. This paper provides a statistical-based methodology to balance the risk of omitting a better solution and the expected computing effort. This methodology not only provides the strong science-based decision making but could also serve as a general tool to be embedded in various single-solution and population-based meta-heuristic studies and provide a cornerstone for further research aiming to provide better search terminating point criteria.

Automated summary

This paper proposes a statistical-based methodology to balance the risk of missing a better solution and expected computing effort when assisting search technique optimizers in making informed decisions about terminating the heuristic search process. The methodology can serve as a general tool for various meta-heuristic studies and lay a foundation for further research on improving search termination criteria.