Research On DNA Microarray Classification Based On Optimized Artificial Bee Colony Algorithm

The rise of DNA microarray technology makes it become a reality for researchers tomeasure the ability of thousands of genes in cells or tissue by performing experiment only once,which provides powerful information and evidence for scientists to obtain the knowledge ofhuman diseases from genome especially the essence of the formation and development of cancer,the diagnosis of disease and relative treatment scheme. The research on data classification forDNA microarray has become a focus in current bioinformatics, and many techniques of datamining and machine learning are widely used in it. However, affected by objective conditionsand human factors, all the data of DNA microarray have the characteristics of high dimensionsand small sample with a large number of invalid and redundant data, which provides newchallenges for traditional techniques such as data mining and machine learning. How to reducethe dimension and scale of data effectively to discover the genes information that determinesdisease becomes a main problem that needs to be solved in the research process of relativealgorithms. Therefore, it becomes the key to design effective methods and classification modelsfor gene selections so as to solve this kind of problems, which is of great importance to improvethe accuracy and efficiency of cancer diagnosis.In this paper, we will discuss how to optimize the design of characteristic of gene selectionsand classification models. Based on artificial bee colony algorithm which has some advantagesin solving combinatorial optimization problem and the classification algorithm of support vectormachine(SVM) that is suitable for solving pattern recognition problem with high dimension andsmall sample, we study the solution to the problem of how to classify the DNA microarray dataand propose gene selection based on optimized artificial bee colony algorithm and anoptimization scheme of synchronization based on support vector machine parameters. All thework is as follows:(1)We optimize artificial bee colony algorithm by combining chaotic theory and catfisheffect. With the help of the chaotic sequence with higher randomness, we enhance thepopulation diversity of the initial bee population diversity and integrate the thought of catfisheffect and chaos theory to derive chaotic catfish bee, which causes effective competition andcoordination to the original bee colony, to break the situation of colony stagnation and toimprove the convergence of the algorithm. (2)We explore the optimal parameters for support vector machine with the help of chaoticcatfish bee colony algorithm. The penalty factor C and kernel function parameter are two keyfactors to dominate the learning ability of support vector machine. We optimize these two kernelparameters by using our chaos catfish bee colony algorithm, which improves the accuracy ofpredicting unknown object.(3)We propose a model with synchronously optimized feature gene selection and SVMparameters based on chaotic catfish bee colony algorithm. Firstly, angle modulation techniquesare used which are able to convert the continuous chaotic catfish bee colony algorithm intodiscrete binary one for feature selections on the original data of genes so as to get effectivesubsets of genes. Secondly, we optimize the parameters of the classifier with the original chaoticcatfish bee colony algorithm on the effective subset to ensure that the most appropriatemodeling parameters can be found for different subsets of genes, consequently, the accuracy ofcancer classification are greatly improved. Finally, simulation experiments are carried out withthe optimized strategies mentioned above on six public data sets of DNA microarray, and theresults show that this scheme has some advantages in the accuracy and efficiency of cancerdiagnosis.