Thesis (Ph.D.)--Chulalongkorn University, 2009

After the human genome project, case-control association studies have been used to extract knowledge of disease association from SNPs. Costs and time in genotyping SNPs are reduced, genotyping SNPs or whole genome is now possible to study the association in a particular disease. However, increasing a number of SNPs affects a number of all possible cases which grows exponentially. A new feature selection and classification called IFGA with BoostMode-SVM is proposed. Two real data sets of case-control association study of Beta-0/Hb E Thalassemia and Crohn's disease from a given set of genotype data are evaluated. The IFGA for feature selection with the BoostMode-SVM classification performs well in both Thalassemia and Crohn's diseases compared with the previous techniques: Optimum Random Forest and CART. We used 6 features as biomarkers for Thalassemia with 71.57% accuracy and 8 features for Crohn's disease with 71.06% accuracy by 10-fold cross validation.