Temel Kavramlar
The proposed Frog-Snake Prey-Predation Relationship Optimization (FSRO) algorithm models the prey-predation relationship between frogs and snakes to solve discrete optimization problems, particularly feature selection in machine learning.
Özet
The content describes a novel swarm intelligence algorithm called Frog-Snake Prey-Predation Relationship Optimization (FSRO) for application to discrete optimization problems, particularly feature selection in machine learning.
The key highlights are:
The algorithm is inspired by the prey-predation relationship between frogs and snakes, modeling the snake's foraging behavior of "search", "approach", and "capture" as well as the frog's characteristic behavior of "standing still, attracting, and then fleeing".
The introduction of evolutionary game theory enables dynamic control of the search process, adjusting the population size of frogs and snakes based on their performance.
Computational experiments were conducted on 26 machine learning datasets to analyze the performance of the proposed FSRO algorithm and compare it with other binary versions of swarm intelligence algorithms.
The results show that FSRO performs competitively with the comparison algorithms on small datasets in terms of fitness value, accuracy, and feature reduction. However, for large high-dimensional datasets, FSRO exhibits slower convergence speed and lower exploitation performance compared to the other algorithms.
The dynamic search control using evolutionary game theory is found to be an effective method, allowing FSRO to achieve a well-balanced search and improve both accuracy and data reduction.
İstatistikler
The number of features in the datasets ranges from 9 to 11,340.
The number of instances ranges from 62 to 5,000.
The number of classes ranges from 2 to 15.
Alıntılar
"The proposed algorithm dynamically controls the search by designing mutation operations with evolutionary stable strategy and adjustment of share of population with replicator dynamics, which are important concepts in EGT."
"Experimental results on 26 datasets from UCI Machine Learning Repository and ASU Feature Selection Repository showed the superior performance of the proposed algorithm compared to the comparison algorithms in terms of the best and standard deviation of fitness value and Accuracy."