<p>The analysis of microarray gene data plays an important role in cancer classification, which can help to detect diseases and cancers. However, the microarray gene data set has a high dimension and there are a large number of redundant and weakly correlated genes, which will cause “curse of dimensionality" when classifying, making the classification problem difficult to deal with. This paper constructs a three-stage gene selection algorithm, which combines the concise PSO algorithm with the fuzzy C-means clustering algorithm based on the intrinsic dimension to solve the problem of gene selection in gene microarray classification, abbreviated as C-ID-HCMPSO. In the first stage (Filtering), the gene selection algorithm uses the symmetric uncertainty of genes to screen the original gene set. In the second stage (Clustering), the fuzzy <i>C</i>-means clustering algorithm is used to cluster the screened genes. During the clustering process, the intrinsic dimension of the original gene data is used to guide the number of clusters. In the third stage (Searching), a concise modified PSO algorithm based on the escape strategy is proposed, and the optimal gene subset is searched by using this swarm intelligence search method. In order to comprehensively evaluate this three-stage gene selection algorithm, experiments have been conducted on 11 binary-class cancer microarray gene datasets and compared with other six algorithms based on PSO. The experimental results show that this algorithm can obtain gene subsets with better classification performance.</p>