Environmental sustainability assessment based on accounting information audit

Area of Science:

• Artificial Intelligence
• Data Science
• Accounting & Auditing

Background:

• Traditional accounting and auditing methods struggle with complex, noisy, and incomplete ESG data.
• Need for advanced feature extraction to improve accuracy and robustness in environmental auditing.

Purpose of the Study:

• To develop a novel feature extraction framework for sustainable accounting and environmental auditing.
• To integrate reinforcement learning-guided steganographic encoding with an EfficientNetV2 backbone.
• To enhance pattern recognition in challenging ESG datasets.

Main Methods:

• Utilized a reinforcement learning-guided steganographic encoding approach.
• Employed an improved EfficientNetV2 backbone with a domain-adaptive multi-branch attention mechanism.
• Integrated a lightweight residual policy network for pattern capture in noisy and imbalanced data.

Main Results:

• Achieved superior classification accuracy, robustness, and explainability compared to state-of-the-art models on ESG datasets.
• Demonstrated an average AUC improvement of 4.7% and a 12.5% reduction in feature redundancy.
• Showcased enhanced performance in privacy-constrained scenarios via steganographic masking.

Conclusions:

• The proposed framework offers significant advancements for sustainable accounting and environmental auditing.
• The method shows strong potential for real-world applications in regulatory auditing and automated compliance.
• Highlights the utility of AI-driven feature extraction for sustainable financial intelligence and ESG reporting.