Accounting Fraud Identification Model Based on Random Forest Algorithm and Light GBM Algorithm
Chujie Sun, Tianle Zhang · 2024
In order to improve the efficiency and accuracy of accounting fraud identification, this paper proposes an accounting fraud identification model based on random forest (RF) and light gradient boosting machine (Light GBM) algorithm. Traditional accounting fraud identification methods often have limitations in the face of complex and diverse financial data, and it is easy to produce a high rate of misjudgment or omission. Machine learning technology, especially ensemble learning algorithm, shows the potential in data-driven fraud detection. Based on the analysis of three types of transactions in the financial fraud detection dataset, this paper constructs an optimized accounting fraud identification model, and makes a comparative experiment with the traditional model. The experimental results show that the proposed optimized model performs well in many performance indexes. In the performance comparison experiment, the prediction time of the proposed optimized model is 0.485 seconds, 0.099 seconds and 0.267 seconds, which shows high real-time detection ability. In the effect comparison experiment, the Area Under Curve of this model is 0.863, 0.899 and 0.974, respectively, and the false positive rate is only 0.015 in the transfer transaction, which shows that this model has high efficiency in identifying fraud and low false positive rate. Therefore, the paper has a certain contribution to the research in the field of accounting fraud detection, which not only verifies the application effect of ensemble learning algorithm in such complex problems, but also provides a reference for model optimization in practical application.