Effectiveness Evaluation of the Probit Model Improved by the Binary Firefly Algorithm in Digital Financial Risk Prediction

In order to address the risk warning of digital finance, it is necessary to deeply explore the underlying patterns to achieve accurate and efficient risk prediction. When using conventional single-class models for data analysis in cases of data imbalance, there are issues with performance instability. This study focuses on shallow feature selection methods for digital financial fraud. To solve the problem of redundant and irrelevant features in digital financial risk data, a Probit model based on the binary firefly algorithm is used to more effectively select initial features from the data. Experimental results show that the binary firefly algorithm with fireworks cooperative evolution converges faster and with higher accuracy compared to other algorithms. In the preliminary feature subset, the optimal classification accuracy is 93.5358%, with an average classification accuracy of 86.2543% and a running time of 5.4883 seconds. The running time has been optimized, and classification accuracy has also been significantly improved. The proposed improved algorithm model can be used to guide financial risk management practices and help financial institutions better identify and prevent financial fraud risks.