A Bayesian approach for identifying multivariate differences between groups
Sverchkov Y, Cooper GF. A Bayesian approach for identifying multivariate differences between groups. In: Proceedings of the International Symposium on Intelligent Data Analysis (2015). Adv Intell Data Anal. 2015 Oct;9385:275-285. Epub 2015 Nov 22. PMCID: 27069983 PMCID: PMC4825814
We present a novel approach to the problem of detecting multivariate statistical differences across groups of data. The need to compare data in a multivariate manner arises naturally in observational studies, randomized trials, comparative effectiveness research, abnormality and anomaly detection scenarios, and other application areas. In such comparisons, it is of interest to identify statistical differences across the groups being compared. The approach we present in this paper addresses this issue by constructing statistical models that describe the groups being compared and using a decomposable Bayesian Dirichlet score of the models to identify variables that behave statistically differently between the groups. In our evaluation, the new method performed significantly better than logistic lasso regression in indentifying differences in a variety of datasets under a variety of conditions.