Corporate biodiversity risk exposure in China: A system-based perspective from natural capital theory using machine and deep learning algorithms

This study examines the firm-level determinants of corporate biodiversity risk exposure by analyzing its associations with operational practices, financial activities, and governance structure. The theoretical framework builds on the system-based perspective of natural capital theory, which views ecosystems as essential assets supporting economic activity. Biodiversity risk is measured using a text analysis method that captures the frequency of biodiversity-related terms in financial reports, based on a structured biodiversity dictionary. The sample includes 1960 publicly listed Chinese firms from 2011 to 2022. To test the proposed framework, we apply machine learning algorithms such as support vector regression, random forest, and K-nearest neighbor, alongside deep learning models including long short-term memory and deep multilayer perceptron. The results show that firms with high resource use, carbon emissions, supply chain concentration, and financial leverage face greater biodiversity risk. In contrast, firms that invest in green innovation, attract institutional investors, and establish environmental governance committees report lower biodiversity exposure. Model validation using normal, cross-validation, and bootstrapping techniques confirms that deep learning models perform better than conventional machine learning in predicting biodiversity risk. The findings offer valuable insights for researchers and policymakers aiming to understand and reduce corporate biodiversity risks in complex industrial and ecological systems.