A Framework for Meta-heuristic Parameter Performance Prediction Using Fitness Landscape Analysis and Machine Learning

The behaviour of an optimization algorithm when attempting to solve a problem depends on the values assigned to its control parameters. For an algorithm to obtain desirable performance, its control parameter values must be chosen based on the current problem. Despite being necessary for optimal performance, selecting appropriate control parameter values is time-consuming, computationally expensive, and challenging. To predict the performance of control parameter configurations in unseen environments, this paper crafts a general framework leveraging machine learning classification and quantitative characteristics of the problem landscape. The proposed framework is extensively explored by training 84 high-accuracy classifiers comprised of multiple sampling methods, fitness types, and binning strategies. Furthermore, a new parameter-reduced particle swarm optimization variant is constructed using the framework to eliminate the computational cost of parameter tuning and yields competitive performance amongst other leading methodologies across 60 unseen benchmark functions.