Gene Data Classification Research Based On The Improved Particle Swarm Optimization And Extreme Learning Machine

Nowadays,the occurrence probability of cancer becomes incrementally higher,which threatens human health.With the development of the human genome project,a large portion of gene expression data has been recognized by DNA microarray technology,where the datasets are usually of high dimension and small sample size.Digging for useful information from the gene expression data is important for the auxiliary tumor diagnosis.Extreme learning machine(ELM)is a tool to solve the complex nonlinear mapping of the input data.It classifies gene data with fast learning speed and avoids convergence toward local minima.However,the classification results are poor and unstable for nonlinear data.In order to obtain a higher accuracy and more stable gene classification algorithm,this paper has researched the ELM algorithm.The main work is:1)A kernel learning machine algorithm based on particle swarm with active operator is proposed.First,we analyze the principle of ELM,introduce kernel functions to the ELM,initialize the given data by adopting KELM,and generate a set of input weights and hidden layer bias.Aiming at solving the problem of the unstable algorithm caused by he random weight assignment in KELM,we utilize the APSO algorithm to optimize the internal weight parameters,and finally obtain the stable APSO-KELM classifier with better classification results.2)An APSO-C-KELM algorithm based on Cholesky decomposition is proposed.For the APSO-KELM algorithm,the kernel function and the particle swarm algorithm with active operator are introduced respectively,which lead to an increment in computational complexity.Besides,the output weight of the KELM algorithm is solved by the basic matrix inverse,which is complex and time-consuming.Therefore,in this part of research,the output weights of KELM are optimized by Cholesky decomposition.The APSO-C-KELM algorithm produces higher accuracy and reduces the running time.