Research On Methods Of Cancer Diagnosis With Genes Based On Machine Learning

Currently, microarray technology which is the joint of life science and information science has become one of the effective methods for human exploration of biomolecule information,and also become one of the most widely used technologies in bioinformatics. Using microarray technology can measure the expression of thousands of genes in one cell simultaneously. Microarray technology provides a premise and technical support for applications such as gene diagnostic and auxiliary therapy.With the rapid development of microarray technology, diagnosis of cancer assisted by gene microarray expression data has attracted more and more attention of researchers. Gene microarray data has the characteristics of high dimensions and less samples. These make the study focus on dimension reduction of high dimensional data and how to use a small number of samples for training and classification, in order to elect genes with classification information and get a higher classification rate.According to such subject,this dissertation, firstly, presented a cancer feature gene selection and classification method based on improved genetic algorithm. In the improved genetic algorithm, the crossover operation is conducted by uniform crossover strategy and mutation operation is conducted by mutation probability nonlinear changing strategy. At the same time, the mutation strategy on optimal individual has been increased according the characteristics of the best subset of feature genes. Uniform crossover strategy has enhanced the global search performance of genetic algorithm. To the later iterations of the genetic algorithm, when individuals gathered near the global optimal solution, increased mutation probability enhanced the local search ability of genetic algorithm effectively, and avoided the premature convergence of genetic algorithm. The mutation strategy on optimal individual made effective use of the global optimal solution of each iteration, which increased the possibility of genetic algorithm finding global optimal solution. The experimental results show that the classification performance of the improved genetic algorithm is obviously performs better than the basic genetic algorithm, and the former also has better robustness.In this paper, the quantum-behaved particle swarm optimization with binary encoding is introduced to the cancer feature gene selection process, and cancer feature gene selection and classification method based on the binary quantum behaved particle swarm optimization algorithm has been put forward.The quantum-behaved particle swarm optimization algorithm with binary encoding is the version of quantum-behaved particle swarm optimization algorithm in binary space, and is one of swarm intelligence algorithms. It is made to be the search engine in the feature space, combined with support vector machine, in order to chose feature gene subset which has strong classification information in the original feature space. The experimental results show that, compared with the particle swarm optimization algorithm and genetic algorithm which are also the search engines, binary quantum-behaved particle swarm optimization algorithm has stronger searching ability and better robustness. It obtained feature gene subset with smaller size and higher accuracy of classification of cancer data set.After analyzing the convergence process of binary quantum-behaved particle swarm algorithm, in order to avoid the premature convergence happened in search process for the optimal solution effectively, variable population size mechanism has been added, so that the algorithm can jump out of the search area in case of premature convergence occurred, and this improved the global search performance of the algorithm. The experimental results show that, the binary quantum-behaved particle swarm algorithm with variable population size mechanism has better search performance compared to the original algorithm, and can find better gene subset in later iterations, and avoided the premature convergence happened effectively.