Statistics and Its Interface

Volume 13 (2020)

Number 1

Bayesian kernel adaptive grouping learning for multi-dimensional datasets

Pages: 127 – 137



Xiaozhou Wang (School of Mathematical Sciences, Shanghai Jiao Tong University, Shanghai, China)

Fangli Dong (School of Mathematical Sciences, SJTU–Yale Joint Center of Biostatistics and Data Science, Shanghai Jiao Tong University, Shanghai, China)


With the development of information technology, a large number of datasets with complex structures, such as multidimensional datasets, need to be processed and analyzed. In this paper we propose a kernel-based statistical learning algorithm, Bayesian Kernel Adaptive Grouping Learning (BKAGL), to provide an innovative solution for the classification problem of multi-dimensional datasets. BKAGL can integrate information from different dimensions adaptively. Meanwhile, we adopt the Bayesian framework which can infer the approximate posterior distributions of parameters. The utilization of grouping features can help find which groups of features have more contributions to the response. Simulation results illustrate that BKAGL outperforms some classical classification methods and the corresponding ungrouped method. The analysis of the electrocardiogram (ECG) and electroencephalography (EEG) datasets shows that BKAGL has the highest classification accuracy and provides explanatory information.


Classifier, multi-dimensional dataset, Bayesian model, adaptiveness, kernel method

2010 Mathematics Subject Classification

62F15, 62H30

The authors were supported by RGC Competitive Earmarked Research Grants, National Basic Research Program of China (973 Program, 2015CB856004) and National Natural Science Foundation of China (11531001).

The authors contributed equally to this work.

Received 1 August 2018

Accepted 15 September 2019

Published 7 November 2019