Identification of Financial Fraud in Listed Companies Based on Bayesian-LightGBM Model

In line with the priorities outlined in the 20th Party Congress report, which emphasizes enhancing financial supervision and mitigating systemic risks, this paper introduces a Bayesian-optimized LightGBM model for improving the accuracy of financial fraud detection in listed companies. The proposed model integrates a hybrid set of indicators, including executive attributes, ESG performance, and innovation capabilities. Using publicly available firm data from 2014 to 2023, the model performs risk identification with a balanced dataset achieved through the KMeans-SMOTE technique to address data skewness. Bayesian optimization is employed to fine-tune the model's hyperparameters, significantly enhancing its predictive accuracy. Experimental results demonstrate a prediction accuracy exceeding 95%, underscoring the model's potential in detecting financial fraud, mitigating risks, and contributing to the stability and sustainable growth of financial markets.