Title: Structural refinement of manually created Bayesian network for prostate cancer diagnosis

Authors: Naveen Kumar Bhimagavni; Thondepu Adilakshmi

Addresses: Department of Computer Science and Engineering, University College of Engineering, Osmania University, Hyderabad, India ' Department of Computer Science and Engineering, Vasavi College of Engineering, Osmania University, Hyderabad, India

Abstract: In general, the structure of a Bayesian network can be learnt from the available data. In some domains like medicine, Bayesian network can be manually created by domain experts and statistical methods can be applied to refine the structure based on the data. As the data is continuously getting evolved in many real-world applications, refinement of the expert network structure is unavoidable. Existing techniques refine the manually constructed Bayesian network either by verifying the relation of a node with the remaining nodes in the network (expert Bayes) or by examining a node only with its parents (MDL principle). In this work, we propose an algorithm that verifies relation of a node only with its non-descendant nodes that are identified with Markov assumption. The proposed algorithm performs small changes to the original network and proves that a smaller number of operations are required to find the best network structure. Maximum likelihood estimation (MLE) is considered as a scoring function to calculate score for each candidate structure and selects the network with the highest score. Manually created Bayesian network has been collected for the widespread disease prostate cancer and proposed algorithm refines the network structure.

Keywords: Bayesian network; prostate cancer; Markov assumption; maximum likelihood estimation; MLE; probabilistic graphical model; PGM; refinement algorithm.

DOI: 10.1504/IJCSYSE.2021.121353

International Journal of Computational Systems Engineering, 2021 Vol.6 No.5, pp.213 - 219

Received: 12 Nov 2020
Accepted: 19 Apr 2021
Published online: 07 Mar 2022 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article