抄録
The survival prediction has become important for patient cancer treatment plan. Previous studies have shown that there is a relation between patient gene expression profile and the survival information. Due to the advancement in microarray technology, high dimensional expressions have been generated. These high dimension microarray data contain many uninformative genes. One of the solutions is to apply a gene selection method to select a small subset of informative genes. A Random Survival Forest method has been selected to perform gene selection for survival prediction. In this method, a score for gene ranking is the key factor to select the smallest possible of gene subsets with lower error rates. Hence, an improved gene selection method using Random Survival Forest has been proposed to obtain gene subsets with a lower error rate. The proposed idea is based on average gene scores for gene ranking. The results showed that the improved gene selection method obtained a lower error rate compared to the standard gene selection methods. Furthermore, the selected subset of genes could assist the predictive ability for survival prediction.
