Classification of bio-data with small data set using additive factor model and SVM

Bio-data, which are obtained from human individuals, have been one of main applications of pattern classification these days. A critical property of bio-data classification is the small number of data in each class due to high cost of obtaining data from each individuals. Since most classification methods are based on the distribution of data in each class, the lack of data can be a main cause of low classification performance of conventional classifiers. To solve this problem, we propose a modified additive factor model for bio-data which has two factors; the individual factor and the environment factor. Under the proposed model, we estimate the distribution of environment factor which gives robust information even in case of small data set. We then define new similarity measures using the information. The similarity measure is applied to nearest neighbor method for classification. We also use the support vector machines (SVM) to find a sophisticated similarity measure. Through computational experiments, we confirm that the proposed model and similarity measure is appropriate enough to show better classification performance compared to conventional similarity measure as well as conventional SVM classifier.