Behavior analysis of carsharing users based on deep belief network and improved two-stage clustering

ABSTRACT

Sustainable development of carsharing plays an important role in reducing the number of private cars and relieving urban parking pressure. Understanding the behavior of users, such as spatiotemporal characteristics of carsharing users, can help operators find high-quality users and provide strong decision support for their service management. To this end, user loyalty based on user behavior is analyzed in this study. Moreover, station distribution of different loyalty types of users is revealed, which is beneficial to optimize the resource distribution. Deep belief network is constructed to predict the future loyalty of users based on order data. An improved two-stage clustering method is designed to mine station usage characteristics of users and determine activity areas of various users. Results show that prediction accuracy of user categories can reach 85% during the six-month observation period. The method of this study can provide a theoretical foundation for the development of a user-centric operation for carsharing.

KEYWORDS:

• Carsharing
• user category prediction
• deep belief network
• spatiotemporal characteristics
• two-stage clustering

Acknowledgments

This work was supported by the Fundamental Research Funds for the Central Universities (2020YJS102) and the National Natural Science Foundation of China (No. 71621001).

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by the National Natural Science Foundation of China [71621001]; the Fundamental Research Funds for the Central Universities [2020YJS102].