改进的BPSO的特征基因选择方法及其在结肠癌检测中的应用研究

Feature Gene Selection Based on Improved Binary Particle Swarm Optimization Algorithm and its Application in Detection of Colon Cancer

为了避免二进制粒子群算法(BPSO)容易陷入局部极值的缺陷,提出了一种改进的二进制粒子群算法(IBP-SO)。该算法在运行过程中引入遗传算法的交叉和变异策略,以便增加种群的多样性,避免粒子的早熟收敛;同时采用免疫算法的疫苗机制,通过合理的疫苗提取、疫苗接种、疫苗选择有效地抑制种群退化的可能。首先采用Wilcoxon秩和检验指标来获得对分类起较大作用的预选特征子集,然后利用IBPSO算法对基因的特征子集和支持向量机(SVM)的参数进行寻优,最后采用IBPSO算法对结肠癌检测问题进行了研究。实验结果表明,该方法可以在较少的特征基因下取得较高精度,且所选的特征基因与结肠癌密切相关,进一步验证了方法的可行性和有效性。

In order to avoid local optimal solution of Binary Particle Swarm Optimization algorithm,an Improved Binary Particle Swarm Optimization （IBPSO） algorithm was presented. In this approach, the crossover and mutational strate- gies are introduced to increase the diversity of populations and avoid the premature-convergence of particles. Vaccine ex- traction, vaccination and immune selection are used to realize the vaccine mechanism to control the population degrada- tion. In order to reduce the features of the tumor, Wilcoxon is used to remove the useless genes. IBPSO algorithm is used to optimize the subset of features and the parameters of Support Vector Machine （SVM）. Finally, this method mentioned above is applied to detect the key genes of colon cancer dataset. The experimental results show that our ap- proach can get higher classification accuracy with smaller size of feature subset than that of some other approaches and the selected genes are proven m be disease-causing. The experimental results also verify the correcmess and effective- ness of our approach.