ReliefF和APSO混合降维算法研究

A hybrid dimension reduction algorithm on ReliefF and APSO

降维与分类一直是机器学习的研究热点,在很多领域有着成功的应用.针对基因数据分类存在特征维数过高、冗余数据和高噪声等问题,现提出一种基于Relief F和自适应粒子群(APSO)优化的混合降维算法.即先通过Relief F和APSO算法选择特征子集,然后使用超限学习机作为评价函数对基因数据进行分类,最后通过循环迭代得到最优的分类精度.实验证明,混合降维算法与已有的算法相比分类精度更高、更稳定,它适用于基因表达数据降维.

Dimensionality reduction and classification are two hot topics in the field of machine learning. We proposed a hybrid feature selection algorithm combining ReliefF and adaptive particle swarm optimization （APSO） for gene data classification, solving the problems of high dimension, redundancy as well as noise. The algorithm extracted the feature subsets by using ReliefF and APSO. The extreme learning machine was used as the evaluation function to classify the gene expression data. The optimized classification accuracy was obtained by recursive substitutions. Experiments show that the proposed hybrid dimensionality reduction algorithm contributes to higher classification accuracy and is more stable than existing algorithms. Consequently, it is an appropriate method for the dimensionality reduction of gene expression data.