A Robust Statistical Framework for the Analysis of the Performances of Stochastic Optimization Algorithms Using the Principles of Severity
Abstract: Meta-heuristic stochastic optimization algorithms are predominantly used to solve complex real-world problems. Numerous new nature-inspired meta-heuristics are being proposed to address various open challenges. Since many heuristics are stochastic, they could yield different solutions to the same problem for different runs. Hence, there is a need for stringent in-depth statistical analysis of the performances of stochastic optimization algorithms. The proposed severity framework enables researchers and practitioners to define application-specific and meaningful performance evaluation metric that evaluates the magnitude of the performance improvement achieved, which is not only of statistical significance but also of practical relevance.
Loading