Article navigation
Purpose

To overcome the limitations of static, checklist-based scoring in capturing the systemic vulnerabilities of infrastructure governance, this study develops an explainable, owner-oriented probabilistic framework to assess and diagnose audit risks in complex expressway projects.

Design/methodology/approach

A mixed-method, data-informed pipeline is established. Audit risk factors are first extracted from historical reports via text mining. Association rule mining (Apriori) is then utilized to discover statistical co-occurrence patterns, which serve as objective dependency candidates. These candidates are refined through audit-process logic and expert screening to optimize a Bayesian Network (BN) topology. Finally, the BN is parameterized using expert elicitation enhanced by fuzzy processing and combination weighting, supporting robust forward, diagnostic, and sensitivity analyses in GeNIe.

Findings

The empirical analysis quantifies a high model-implied audit risk exposure (87.08%), demonstrating that audit risks are driven by complex, multi-domain interactions rather than isolated failures. Crucially, diagnostic (backward) inference pinpointed specific priority drivers under adverse audit scenarios, including unreasonable approval of budget estimates (81.52%), incorrect interest calculation (71.54%), and non-compliant construction management fees (62.26%). Furthermore, the network successfully uncovered latent risk pathways, such as the cascaded impact of inaccurate traffic volume forecasting, which are frequently overlooked in traditional retrospective audits. Sensitivity analysis identified the highest-leverage control nodes, directly translating probabilistic inference into targeted resource allocation strategies.

Originality/value

The primary innovation of this study lies in its methodological integration: it uses statistical data mining (Apriori) to objectively propose BN dependency cues, thereby mitigating the subjectivity of traditional expert-driven network structuring while avoiding the “black-box” opacity of pure machine learning models. This research scientifically justifies the shift from reactive, isolated compliance checks to a dynamic, transparent early-warning mechanism, offering actionable, explainable insights for owner-side infrastructure audit governance.

Licensed re-use rights only
You do not currently have access to this content.
Don't already have an account? Register

Purchased this content as a guest? Enter your email address to restore access.

Pay-Per-View Access
$41.00
Rental

or Create an Account

Close subscription notice
Close access options