AI-Driven Financial Fraud Auditing and Evidence Chain Construction in Smart Manufacturing

INTRODUCTION: In smart manufacturing enterprises, procurement, production, warehousing, sales, and financial settlement data are distributed across multiple business systems. Financial fraud increasingly involves cross-entity collusion, nested business chains, and fragmented evidence, making it difficult for traditional sampling audits to simultaneously ensure accurate risk identification, efficient verification, and traceable evidence support. OBJECTIVES: To address fragmented cross-system associations among manufacturing audit data, difficulties in tracing multi-hop fraud propagation across business entities, and insufficient evidence support for model-generated risk alerts, this study develops an AI-driven financial fraud auditing framework tailored to smart manufacturing processes. The framework links fraud risk identification, dynamic audit-path optimization, and verifiable evidence-chain construction within a unified closed-loop procedure. METHODS: Multi-source heterogeneous audit samples are constructed through entity matching, temporal alignment, and data-quality filtering. Business process constraints are introduced to encode inconsistencies in transaction amounts, material quantities, event timing, and entity relationships. A heterogeneous graph neural network jointly identifies high-risk entities, anomalous transactions, and fraud categories while tracing cross-entity risk propagation paths. The audit process is further formulated as a sequential decision-making task, in which deep reinforcement learning dynamically selects verification actions according to risk information gain, execution cost, and evidence sufficiency. Original records, risk relationships, audit actions, and review statuses are subsequently organized into a structured and traceable evidence chain. RESULTS:Simulation experiments show that the proposed method achieves an F1 score of 92.7% and an Area Under the Curve (AUC) of 98.4% for fraud detection, together with a 90.7% recall rate for risk propagation paths. Evidence completeness reaches 95.4%, while the average number of audit steps required to establish a complete evidence chain decreases to 6.9. Direct real-world testing was further conducted on three anonymized manufacturing enterprises containing 715,220 raw cross-system records and 3,660 independently audited business chains. With all trained parameters and decision thresholds fixed, the framework identified 271 of 307 confirmed anomalous chains and achieved a pooled Precision of 89.74%, Recall of 88.27%, F1 score of 89.00%, AUC of 0.956, and false alarm rate of 0.92%. Risk-propagation-path recall and evidence completeness reached 86.64% and 92.18%, respectively, while the average number of audit steps was 7.5. CONCLUSION: Integrating business-process-constrained heterogeneous graph reasoning, risk-feedback-driven audit decision-making, and structured evidence verification enables complex financial fraud to be identified, localized, and substantiated within a unified audit workflow. The proposed framework extends conventional fraud detection from isolated risk classification toward traceable cross-entity reasoning and evidence-supported audit decision-making in smart manufacturing environments.

Authors

Institutions

Publication Details

Journal
ICST Transactions on Scalable Information Systems
Published
2026-11-20
DOI
https://doi.org/10.4108/eetsis.14854
Primary Topic
Benford’s Law and Fraud Detection
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

AI-Driven Financial Fraud Auditing and Evidence Chain Construction in Smart Manufacturing

Zhaojun Hou
1 links
ICST Transactions on Scalable Information Systems
Benford’s Law and Fraud Detection
article

AI-Driven Financial Fraud Auditing and Evidence Chain Construction in Smart Manufacturing

Zhaojun Hou
article en

Abstract

INTRODUCTION: In smart manufacturing enterprises, procurement, production, warehousing, sales, and financial settlement data are distributed across multiple business systems. Financial fraud increasingly involves cross-entity collusion, nested business chains, and fragmented evidence, making it difficult for traditional sampling audits to simultaneously ensure accurate risk identification, efficient verification, and traceable evidence support. OBJECTIVES: To address fragmented cross-system associations among manufacturing audit data, difficulties in tracing multi-hop fraud propagation across business entities, and insufficient evidence support for model-generated risk alerts, this study develops an AI-driven financial fraud auditing framework tailored to smart manufacturing processes. The framework links fraud risk identification, dynamic audit-path optimization, and verifiable evidence-chain construction within a unified closed-loop procedure. METHODS: Multi-source heterogeneous audit samples are constructed through entity matching, temporal alignment, and data-quality filtering. Business process constraints are introduced to encode inconsistencies in transaction amounts, material quantities, event timing, and entity relationships. A heterogeneous graph neural network jointly identifies high-risk entities, anomalous transactions, and fraud categories while tracing cross-entity risk propagation paths. The audit process is further formulated as a sequential decision-making task, in which deep reinforcement learning dynamically selects verification actions according to risk information gain, execution cost, and evidence sufficiency. Original records, risk relationships, audit actions, and review statuses are subsequently organized into a structured and traceable evidence chain. RESULTS:Simulation experiments show that the proposed method achieves an F1 score of 92.7% and an Area Under the Curve (AUC) of 98.4% for fraud detection, together with a 90.7% recall rate for risk propagation paths. Evidence completeness reaches 95.4%, while the average number of audit steps required to establish a complete evidence chain decreases to 6.9. Direct real-world testing was further conducted on three anonymized manufacturing enterprises containing 715,220 raw cross-system records and 3,660 independently audited business chains. With all trained parameters and decision thresholds fixed, the framework identified 271 of 307 confirmed anomalous chains and achieved a pooled Precision of 89.74%, Recall of 88.27%, F1 score of 89.00%, AUC of 0.956, and false alarm rate of 0.92%. Risk-propagation-path recall and evidence completeness reached 86.64% and 92.18%, respectively, while the average number of audit steps was 7.5. CONCLUSION: Integrating business-process-constrained heterogeneous graph reasoning, risk-feedback-driven audit decision-making, and structured evidence verification enables complex financial fraud to be identified, localized, and substantiated within a unified audit workflow. The proposed framework extends conventional fraud detection from isolated risk classification toward traceable cross-entity reasoning and evidence-supported audit decision-making in smart manufacturing environments.

ICST Transactions on Scalable Information SystemsVol. 13(8)
Shandong Normal University (CN)
Benford’s Law and Fraud Detection
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon