Liming Yang1, Yongping Gao2, Qun Sun3
CMES-Computer Modeling in Engineering & Sciences, Vol.104, No.6, pp. 493-506, 2015, DOI:10.3970/cmes.2015.104.493
Abstract Minimax probability machine (MPM) has been recently proposed and shown its advantage in pattern recognition. In this paper, we present a new minimax probabilistic approach (MPA),which can provide an explicit lower bound on prediction accuracy. Applying the Chebyshev-Cantelli inequality, the MPA is posed as a second order cone program formulation and solved effectively. Following that, this method is exploited directly to recognize the purity of hybrid seeds using near-infrared spectroscopic data. Experimental results in different spectral regions show that the proposed MPA is competitive with the existing minimax probability machine and support vector machine in generalization, while requires less computational… More >