Early Warning of Supply Chain Risks Based on RF-GA-BPNN Algorithm

Supply chain systems are constantly facing multiple risks and challenges from both internal and external environments. In the existing research, early warning algorithms of supply chain risks are still insufficient in terms of indicator selection and threshold optimization. In order to further improve the early warning ability of supply chain risks, this paper establishes an early warning model based on the RF-GA-BPNN algorithm focusing on algorithm fusion optimization and its warning effectiveness. The model organically combines the characteristics and advantages of multiple algorithms, including random forest, genetic algorithm, BP neural networks, to improve the prediction effect of the BP neural network by means of indicator feature importance screening and initial parameter optimization. The model is trained and tested by using a risk early warning indicator dataset from 3,309 listed companies in Chinese A-share market. The prediction results show that the RF-GA-BPNN algorithm achieves an early warning accuracy of 96.50% after training with 300 sets of random sample data. The proposed model based on the RF-GA-BPNN algorithm has excellent learning and early warning capabilities. The prediction results provide numerical references for the initial judgment of supply chain risk levels and the implementation of risk resistance measures.