A Mining Algorithm to Improve LSTM for Predicting Customer Churn in Railway Freight Traffic

Fangcan ZHAO, Baotian DONG, Hongqin PAN, Anqi SHI

Abstract

Railway freight is at risk of losing customers due to intense competition in the market. Effectively managing potential customer loss is a long-term problem for railway freight. Big data technology has been widely researched and applied in recent years, and using data mining techniques to fully extract information from railway freight ticket data and discover potential customer loss is a research topic. In this study, a mining algorithm is proposed to improve the accuracy of predicting customer churn in railway freight using Long Short-Term Memory (LSTM) network. The algorithm involves three steps: constructing the time series of railway freight volume, predicting customer churn trends using a modified LSTM model, and analyzing the characteristics of customers with different degrees of loss and the characteristics of transportation demand of lost and potential lost customers. Experimental verification was conducted using customer freight ticket data from the Shanghai Railway Bureau in China and compared the proposed modified LSTM model was compared with other commonly used machine learning algorithms for churn prediction. The results showed that the present algorithm demonstrated good accuracy and adaptability. The study proposed in this paper enriches the theoretical basis of railway freight customer management, provides an effective method for predicting customer loss in railway freight, and offers technical support for railway freight management practices.

Keywords

Rail freight, Customer churn prediction, Big Data, LSTM, Data mining, Transportation demand, Customer management.

View full article