A Normal-Gamma Dirichlet Process Mixture Model
share
We propose a Dirichlet process mixture (DPM) for prediction and cluster-wise variable selection, based on a Normal-Gamma baseline distribution on the linear regression coefficients, and which gives rise to strong posterior consistency. A simulation study and real data application showed that in terms of predictive and variable selection accuracy, the model tended to outperform the standard DPM model assigned a normal prior with no variable selection. Software code is provided in the Supplementary Information.
READ FULL TEXT