Skip to Main Content
This paper presents two genetic algorithm (GA) based hybrid approaches for the prediction of tumor outcomes based on gene expression data. One approach is the hybrid GA and K-medoids for grouping genes based on the commonly used distance similarity. The goal of grouping genes here is to choose some top-ranked representatives from each cluster for gene dimensionality reduction. The second proposed approach is the hybrid GA and Support Vector Machines (SVM) for selecting marker genes and classifying tumor types or predicting treatment outcomes. These two hybrid approaches have been applied to public brain cancer datasets, and the experimental results are compared with those given in a 2001 paper published in the Nature. The final prediction accuracies are found to be superior both for tumor class prediction and treatment outcome prediction.