Feature Extraction of Colon Cancer´s Gene

In this paper, the gene features associated with colon cancer were extracted from the given colon cancer gene expression data by reasonable statistical methods and other mathematical methods. Firstly, the majority(80%-90%), of unrelated genes of which the linear normalized distance was below 0.2 were removed using Bhattacharyya weighted distance, then the weight of the corresponding trait genes and the corresponding threshold value were derived using the linear kernel SVM training feature gene subset. We found the best combination of genes(2-60) and picked out the optimal combination of genes with the lowest error rate as the ultimate combination of genes(5-7). Finally, we tested the optimal combination of genes on the test set and the error rate was below 3%, indicating high precision.