A nonparametric Bayesian prediction approach for modelling small data

In the 21st century’s big data era, the small data snags still exist in many areas which causes modelling and evaluation are hard to make. Small dataset is insufficient to generate a reliable prediction especially in the small area estimation domain. This paper presents a Bayesian nonparametric prediction
framework to models which are dealing with smaller data sets and/or give poor predictions. This approach uses a robust Gaussian model in weight-space notion and drive the prediction distribution of the future responses. Results revealed that the prediction distribution of a set of futures responses is conditional on a set of observed data and depends on the degree of the spline. It also provides an
empirical illustration and demonstrated that the prediction outcomes depend on the realised responses only through the observations in design matrix and the sample residual sum of squares and products matrices with the Kronecker product.