Application Of Improved PSO Algorithm And ELM In Classification Of Genetic Data

Nowdays,the incidences of cancer appear more frequently,and the threats to humans are increasing.With the rapid development of biomedicine,the technology of gene classification is gradually matured.In the field of data mining,biomedical images have appeared and the classification and recognition of malignant tumor diseases such as cancer have been accomplished by using machine learning algorithms.The gene expression data provides a reliable theoretical base for the diagnosis and treatment of diseases by analyzing the changes of genes as well as the interrelationship of genes.Extreme Learning Machine(ELM),which is a learning algorithm with fast classification speed,has attracted increasing number of researchers and has been widely used in many fields.ELM is able to approximate the complex non-linear mapping,classify the genetic data efficiently,and avoid local minimum/maximum.However,when applied to non-linear data,the classification accuracy falls due to the random setting of input weight and hidden layer bias.To solve this problem,Particle Swarm Optimization(PSO)is introduced to optimize the parameters.In this study,we focus on the following topics:(1)A new Kernel Extreme Learning Machine is proposed,which is based on the Kernel Particle Swarm Optimization algorithm(DPSO-WKELM).The solution of ELM randomly generates the input layer weights in the training process,finds hidden layer node bias on the stability of the model and produces reasonable classification accuracy with better stability and generalization.In this paper,we use the Detecting Particle Swarm Optimization(DPSO)algorithm to optimize the input weights and hidden layer bias of the ELM,.We add the detection particles,and optimize the individual particles to the ordinary particles as their current position for subsequent iterations.The convergence rate and the convergence precision of the PSO in the global optimization are effectively improved.The experimental results show that DPSO-WKELM can obtain higher classification precision in genetic data classification.(2)The parameter randomization of ELM is not conducive to the generalization of the model.The adaptive chaos particle swarm optimization(ACPSO-ELM)is proposed to optimize the parameters of the ELM.In this paper,the inertia factor is added to the particle velocity update formula,PSO algorithm can solve the problem of premature convergence and add chaos sequence to the particle swarm,which makes the particle escapes whenever it is inert..The experimental results show that the adaptive chaotic sequence can improve the optimization process of the particle as well as the precision of the genetic algorithm.