A New-type Deep Learning Model Based on Shapley Regulation for Containerized Freight Index Prediction

Abstract

In this study, we have crafted an innovative methodology that represents a groundbreaking synthesis of deep learning techniques with cooperative game theory. In this study, we use the accuracy of data prediction by different LSTM models as a measurement index and assign different LSTM models corresponding weights through the Shapley value calculation method to construct a more accurate predictive analysis model. We use this improved Shapley regulation model to calibrate a long short-term memory (LSTM) neural network by using historical freight data to predict the China Container Freight Index (CCFI), the leading export container freight index commonly used in China. Afterward, it is found that the neural networks calibrated in this way reduce their prediction bias in terms of mean absolute percentage error (MAPE), mean absolute error (MAE), root mean square error (RMSE), and mean square error (MSE) to improve prediction accuracy.