We propose a best predicted curve classification (BPCC) criterion for classifying the curve data. The data are viewed as realizations of a mixture of stochastic processes and each subprocess corresponds to a known class. Under the assumption that all the groups have different mean functions and eigenspaces, an observed curve is classified into the best predicted class by minimizing the distance between the observed and predicted curves via subspace projection among all classes based on the functional principal component analysis (FPCA) model. The BPCC approach accounts for both the means and the modes of variation differentials among classes while other classical functional classification methods consider the differences in mean functions only. Practical performance of the proposed method is demonstrated through simulation studies and a real data example of matrix assisted laser desorption (MALDI) mass spectrometry (MS) data. The proposed method is also compared with other multivariate and functional classification approaches. Overall, the BPCC method outperforms the others when the mean functions and the eigenspaces among classes are significantly distinct. For classifying the MALDI MS data, we found that functional classification methods perform better than multivariate data approaches, and the dimension reduction via FPCA is advantageous to improving the accuracy of classification.
International Journal of Intelligent Technologies and Applied Statistics 3(4), pp.383-399