Improved Multiobjective Artificial Bee Colony Algorithm And Its Application In Feature Selection

Swarm intelligence algorithm is a kind of algorithm that simulates the biological evolution in nature.Simulating original distribution of species by randomly initializing individuals,simulating the natural evolution of the species through random search and selective preservation of the offspring,so as to realize the evolution of the population of the algorithm and find the optimal solution.Artificial bee colony algorithm is one of the swarm intelligence algorithm,which is a new kind of swarm intelligence heuristic algorithm realizing random search by imitating employeed bees,realizing local search by imitating onlooker bees,and finally realizing the reinitialization of population by imitating scout bees,the three processes are executed in order to achieve the goal of algorithm optimization.Due to its simpleness,simpleness and quickness,artificial bee colony algorithm has been widely used in numerical optimization of functions,optimization of manufacturing processes,engineering design and chemical engineering.Howerer,the artificial bee colony algorithm still has the defects of slow convergence and easily falling into the local optimum in the process of evolution and optimization as a new algorithm,which affects its performance in solving practical problems to a certain extent.Therefore,studying the internal principle of artificial bee colony algorithm,improving it and extending it to multi-objective area,is of great value and practical significance for solving practical problems.In this paper,by studying the existing improved artificial bee colony algorithm,aiming at the existing problems in the algorithm and combining with the basic principle of swarm intelligence algorithm,several improved schemes are put forward,latter,the performance of this algorithm is tested on the basic test function,and extends the improved Artificial bee colony algorithm to multi-target area and explores its application in the field of feature selection.This article is divided into two parts:1.In order to solve the problem of insufficient population convergence pressure in later stage caused by employeed bee phase in the original multi-objective artificial bee colony algorithm using pareto-dominated selection mechanism,the idea that guiding evolution using Knee Points is proposed and applicating it to the employeed and onlooker bee phase to accelerate algorithm converges,in the later stage of the algorithm,the idea can provide sufficient convergence pressure.Based on the above ideas,this paper proposes an improved multi-objective artificial bee colony algorithm based Knee Points(KnMOABC).In this algorithm,an mechanism for adaptively identifying Knee Points is first designed.Each neighborhood is adaptively divided in every pareto front in population and the point with the farthest distance in the neighborhood is regard as Knee Point.At the same time,An adaptive algorithm that determines the number of Knee Points based on the Implementation of the algorithm is proposed.In the early stage of the algorithm,the number of Knee Points only needs to be kept at a relatively low level as the population itself has sufficient evolutionary pressure so that it can play an assisting evolutionary role in guiding the population evolution.In the later stage of algorithm implementation,Because most of the individuals in the population are distributed on the first front and are non-dominated with each other,the selection criteria based solely on the Pareto domination and crowding distances can not provide enough convergence pressure in the later stage in evolution,finally increasing the number of Knee points to provides enough pressure in employeed and onlooker bee phase helping the population evolving to the ideal Pareto front.2.In the classical multi-objective bee colony algorithm,the neighbor-based employeed,onlooker bee phase and completely random scout-bee.mechanism are adopted.This mechanism is prone to oversize variation in the early stage of the algorithm evolution,miss the optimal value point,and in the later stage of the algorithm,the step is too small and the local optimal is caught.In order to solve this problem,a method of adaptively adjusting the step size is proposed in this paper.The variation step for every individual is calculated by number of invalid repetitions,achieving the purpose of automatically adjusting the search neighborhood according to the quality of each nectar source.In the later stage in the algorithm,the regenerated individuals can easily far away from the current population in the late of the algorithm by the completely random scout bee phase,which seriously hinders the convergence speed of the algorithm.In view of this problem,this paper proposes a Gaussian initialization method based on the optimal point,determining the regenerated individual according to the average position of the population.3.This paper presents a method that using the Knee Points-based improved multi-objective artificial bee colony algorithm in solving the problem of feature selection problem,taking the advantaging of the features that rapid convergence and strong searching capability to solve the feature selection,one of the NP-Hard problems.The performance of the feature selection method is tested on the eleven UCI datasets with other classical multi-objective evolutionary algorithms using Matlab simulation.The simulation results show that the proposed algorithm has a good search result for solving the feature selection problem,showing better search performance.