Query Specific Pruning Strategies in Bayesian Networks for Inference Efficiency

Loading...
Thumbnail Image

Journal Title

Journal ISSN

Volume Title

Publisher

University of the Witwatersrand, Johannesburg

Abstract

Bayesian networks (BNs) serve as a powerful tool for encoding probabilistic relationships among variables and facilitating inferential tasks. However, the complexity and size of these networks can pose significant challenges, particularly in scenarios demanding real-time decision-making. This research introduces query-specific pruning strategies designed to improve the efficiency of inference processes in Bayesian networks. By selectively pruning nodes and edges that are irrelevant to a given query, these strategies considerably reduce the computational overhead without compromising the accuracy of the inference results. Our approach uniquely used several pruning techniques, each tailored to different aspects of the network’s structure and the specifics of the query and evidence presented. These include basic pruning, refined pruning, V-structure pruning, and community-based pruning, all of which have been systematically evaluated to assess their impact on inference cost and accuracy. The effectiveness of these methods is demonstrated through a series of experiments conducted on synthetic networks as well as on real world data. The results reveal that query-specific pruning significantly outperforms traditional inference methods namely variable elimination, especially in large-scale networks, by efficiently managing the scope of the probabilistic inference task. This not only accelerates the inference process but also retains high fidelity in the results, thus supporting robust decision-making in critical applications such as healthcare. Through this research, we aim to provide a methodological framework that can be adapted to various domains, promoting more efficient use of Bayesian networks in practical settings.

Description

A dissertation submitted in partial fulfillment of the requirements for the degree of Masters in Computer Science, to the Faculty of Science, School of Computer Science and Applied Mathematics, University of the Witwatersrand, Johannesburg, 2025

Citation

Chirkoot, Mikhail. (2025). Query Specific Pruning Strategies in Bayesian Networks for Inference Efficiency. [Master's dissertation, University of the Witwatersrand, Johannesburg]. WIReDSpace. https://hdl.handle.net/10539/50203

Endorsement

Review

Supplemented By

Referenced By