In this paper, a probabilistic framework for ordinal prediction is proposed, which can be used in modeling ordinal regression. A sparse Bayesian treatment for ordinal regression is given by us, in which an automatic relevance determination prior over weights is used. The inference techniques based on Laplace approximation is adopted for model selection. By this approach accurate prediction models can be derived, which typically utilize dramatically fewer basis functions than the comparable supported vector based and Gaussian process based approaches while offering a number of additional advantages. Experimental results on the real-world data set show that the generalization performance competitive with support vector-based method and Gaussian process-based method.
展开▼
