An Efficient Approach to Learn an Effective Hierarchy of a Set of OOBN Classes

Day by day, Bayesian networks are getting popular for solving real-life problems. However, it is difficult to build Bayesian decision networks (BNs) to solve large scale real world problems. Using object-oriented Bayesian networks (OOBNs) is one strategy to deal with the scalability issue. OOBNs make it possible by providing researchers with the facility to design classes and build models with a modular and hierarchical architecture, which increases reuse and maintenance facilities. Sharing properties down the hierarchy of classes, known as “inheritance” in OO-paradigm, is a key idea to increase the reusability and tackle scalability issue. It means that one can share or reuse components and behaviors of an entity known as object or class. Previously, a framework of OOBN was proposed to contain inheritance and all other aspects of OO-paradigm. Recently, in 2022, an extension was proposed to learn hierarchy of OOBN classes. However, such an extension is still suboptimal. In this paper, we identify some scopes to improve the learning technique. We propose and implement a new algorithm and then analyze it empirically and asymptotically. We use both synthetic and real-world data in the empirical analysis. The analysis shows that our proposed algorithm is more effective and efficient, especially in terms of reusability.