Heterogeneous Ensemble Model Based Financial Statement Fraud Detection of Listed Companies

Financial statement fraud in listed companies has become increasingly complex, systematic, and difficult to detect, posing significant challenges for financial supervision. This paper presents an intelligent detection approach using a heterogeneous ensemble model to identify fraud in listed companies, focusing on the Chinese manufacturing industry. The study involved the selection and preprocessing of financial report samples, addressing sample imbalances, and constructing an ensemble framework combining XGBoost, K-nearest neighbors, Random Forest, and Support Vector Machine as primary classifiers, with Logistic Regression as the secondary classifier. Cross-validation results showed significant improvements in precision, recall, F1 score, and accuracy compared to individual models, with precision increasing by 7.22%, recall by 2.33%, F1 score by 4.53%, and accuracy by 1.76%. The findings suggest that this model offers valuable decision-making support for detecting financial fraud in listed companies.