An approach for knowledge acquisition from a survey data by conducting Bayesian network modeling, adopting the robust coplot method


JOURNAL OF APPLIED STATISTICS, vol.49, no.16, pp.4069-4096, 2022 (SCI-Expanded) identifier identifier identifier

  • Publication Type: Article / Article
  • Volume: 49 Issue: 16
  • Publication Date: 2022
  • Doi Number: 10.1080/02664763.2021.1971631
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, ABI/INFORM, Aerospace Database, Business Source Elite, Business Source Premier, CAB Abstracts, Veterinary Science Database, zbMATH
  • Page Numbers: pp.4069-4096
  • Keywords: Bayesian network, structural learning, graphical visualization, robust coplot analysis, variable elimination, VARIABLE SELECTION
  • Hacettepe University Affiliated: Yes


This study proposes a methodological approach for extracting useful knowledge from survey data by performing Bayesian network (BN) modeling and adopting the robust coplot analysis results as prior knowledge about association patterns hidden in the data. By addressing the issue of BN construction when the expert knowledge is limited/not available, this proposed approach facilitates the modeling of large data sets describing numerously observed and latent variables. By answering the question of which node(s)/link(s) should be retained or discarded from a BN, we aim to determine a compact model of variables while considering the desired properties of data. The proposed method steps are explained on real data extracted from Turkey Demographic and Health Survey. First, a BN structure is created, which is based solely on the judgment of the analyst. Then the coplot results are employed to update the BN structure and the model parameters are updated using the updated structure and data. Loss scores of the BNs are used to ensure the success of the updated BN that inherits knowledge from coplot.