Abstract
Recent measurement technologies in molecular biology allow us to obtain high-throughput human genomic information such as single nucleotide polymorphisms (SNPs). Many epidemiological studies indicate that these SNP information are useful for predicting the risk of common diseases such as stroke, diabetes or hyper-tension, and several attempts have been done for developing predictive risk models using machine learning approach. In order to improve such disease risk prediction models, we introduce in this paper a novel machine learning approach that can efficiently identify a subset of SNP combinations that are relevant to the risk of these common diseases.
