Study on Performance Evaluation of Fresh Agricultural Supply Chain Based on BP Neural Network

Abstract

A BP neural network is an algorithm for forward multi-layer backpropagation learning. Its basic idea is forward propagation and error backpropagation in the learning process. At present, the most common result of a BP neural network is a three-layer structure. In view of the performance evaluation of fresh agricultural supply chain, this paper proposes a performance evaluation method based on BP neural network. First, three first-level indicators and six more detailed second-level indicators are set up according to the characteristics of the data, and the BP neural network is trained and tested by cross-checking method. The BP neural network after training was evaluated by using the confusion matrix, accuracy and recall rate, MMC, ROC curve, and AUC value. It was found that in the confusion matrix output by BP neural network, TP values of the three first-level indicators were all large, while the accuracy and recall rate, MMC, ROC curve, and AUC values were all high. The values of 0.958, 0.678, and 0.588 respectively indicate that BPNN has good reliability and prediction accuracy. This paper further compares the BP neural network, decision tree model, SVM, and various evaluation results of ARIMA. The BP neural network is second only to ARIMA in accuracy and recall rate, and improves MCC and AUC values by 10.54% and 14.05% compared with ARIMA, with the best comprehensive performance. Meanwhile, with the increase of data volume, compared with the other three models, BP neural network has more advantages on AUC and has stronger evaluation authenticity and reliability in the big data environment.