Original scientific paper

https://doi.org/10.17559/TV-20250514002665

Exploration of ESG Audit Adaptive Decision-Making and Anomaly Analysis Driven by Reinforcement Learning in Artificial Intelligence

Jie Wang ; School of business, Anhui University of Technology, Maanshan 243032, China
Hongquan Li ; School of business, Anhui University of Technology, Maanshan 243032, China *
Feiyang Yu ; School of business, Anhui University of Technology, Maanshan 243032, China

* Corresponding author.

Abstract

The dynamic nature of Environmental, Social, and Governance (ESG) audit strategies, particularly in responding to evolving environmental regulations and shifting corporate sustainability practices, necessitates robust methodological innovation. Reinforcement learning (RL) presents a transformative pathway for refining ESG audit processes through continuous interaction with environmental performance data, regulatory updates, and real-time ecological compliance feedback. Nevertheless, the application of RL to optimize ESG auditing remains significantly underexplored. Addressing this gap, our study develops an RL-driven model designed to strategically recalibrate ESG auditing mechanisms, with enhanced focus on responsiveness to emerging environmental compliance requirements and ecological risk factors. We adopt a dual-method research framework integrating theoretical and empirical approaches. The theoretical investigation establishes the structural compatibility between RL algorithms and environmental ESG audit optimization, ensuring alignment with the complexities of sustainability decision-making. Empirical validation employs: (1) large-scale simulations using synthetic corporate environmental datasets to evaluate model performance across diverse operational scenarios, and (2) real-world applications to quantify the model's efficacy in improving ecological audit efficiency, mitigating environmental compliance risks, and addressing the challenges of modern sustainability auditing. Results demonstrate that the RL-driven model outperforms conventional methods in adapting to environmental data variability, achieving a 40% increase in resource efficiency and a 30% improvement in predicting ecological compliance risks. Practical implementations further reveal a 25% reduction in audit cycle duration and significantly fewer errors in environmental disclosure assessments. These findings highlight RL's potential to revolutionize environmentally focused ESG auditing while underscoring ongoing challenges in data reliability and model interpretability for sustainability applications.

Keywords

artificial intelligence; deep Q-network (DQN); ESG auditing; Markov decision process; reinforcement learning

Hrčak ID:

342627

URI

https://hrcak.srce.hr/342627

Publication date:

31.12.2025.

Visits: 556 *