Payment Behavior Prediction and Statistical Analysis for Shared Parking Lots

Qingyu Xu, Feng Zhang, Mingde Zhang, Jidong Zhai, Jiazao Lin, Haidi Liu, Xiaoyong DuNPC 2020 · OpenReview

TL;DR

This paper tackles the challenge of predicting payment behavior in shared parking lots, a growing segment of the sharing economy. By analyzing real-world parking data, the authors develop a decision tree model that accurately predicts whether and when users will pay for parking, offering a practical solution for improving service and reducing payment issues.

What Problem Does This Paper Address?

The rapid growth of the sharing economy, particularly in urban areas with increasing car ownership, has led to the proliferation of intelligent shared parking lots. A significant operational challenge for these services is "untimely payment behavior," where users delay or forget to pay for parking. This not only impacts revenue but also complicates operations. The authors aim to address this by predicting user payment behavior, specifically focusing on when users will pay their parking bills. This prediction can enable proactive interventions, such as payment reminders, to mitigate delayed payments and enhance user experience. The problem is compounded by the fact that many users have limited historical parking data, making traditional prediction methods difficult.

Methods

The core of the paper's methodology involves a two-pronged approach: statistical analysis of influencing factors and predictive modeling.

  1. Feature Analysis: The authors first analyze a dataset of parking records to identify factors that influence payment behavior. They categorize these features into:
  2. Categorical Features: Year, month, user identity (membership status).
  3. Numerical Features: Order amount minus deposit, total number of parkings, parking duration, and average payment duration. They explore the relationships between these features and payment behavior, noting trends like an increase in orders to be paid over time and a dip during the COVID-19 pandemic. They also identify that users with unpaid records often have very few parking instances.

  4. Predictive Modeling: Based on the insights from the feature analysis, the authors employ a Decision Tree model, specifically the CART (Classification And Regression Trees) algorithm. This model is chosen for its simplicity and effectiveness in classification tasks. They train and evaluate the model using 10-fold cross-validation. The model is designed to predict the payment type (timely, late, or arrears) and, by extension, the payment time.

Datasets

The study utilizes a dataset of 29,733 real parking records provided by Huaching Tech, a smart parking company in China. This dataset spans 14 months, from April 9th, 2019, to June 29th, 2020. The records were collected from a randomly selected parking lot in Hangzhou, China. The dataset is divided into three categories based on payment duration:

  • Arrears: Payment duration > 3 days.
  • Late Payments: 1 day < Payment duration <= 3 days.
  • Timely Payments: Payment duration < 1 hour.

Key Insights and Ideas

  • Payment Behavior is Predictable: Even with limited historical data for many users, payment behavior in shared parking lots can be predicted with reasonable accuracy.
  • Temporal and User Factors Matter: The time of year (month, year) and user characteristics (like membership) are significant categorical features influencing payment.
  • Parking Duration as a Key Numerical Feature: The duration of parking is a crucial numerical factor, with records under 5 minutes being treated as noise.
  • The "New User" Problem: Users with very few parking records (one or two) are more likely to exhibit delayed or unpaid behavior, highlighting a challenge in onboarding new users.
  • COVID-19 Impact: The data shows a notable decrease in payments in February 2020, directly attributed to the COVID-19 pandemic, underscoring the influence of external events.

Key Results

The decision tree model achieved an accuracy of 79.73% in predicting payment behavior. The study also found that:

  • 80.26% of cases were timely payments.
  • 2.96% were late payments.
  • 16.79% were arrears. This distribution suggests that while delayed payments are a concern, the majority of users pay on time. The model's strength lies in its ability to identify potentially unpayable orders, which is critical for shared parking operations.

Limitations and Future Work

Limitations: * The paper acknowledges that predicting payment behavior with limited information (few parking records per user) poses significant difficulties. * The study focuses on a single parking lot in Hangzhou, China, which might limit the generalizability of findings to other regions or types of parking services.

Future Work: * The authors suggest that their method is effective for identifying unpayable orders, implying that future work could focus on developing more sophisticated intervention strategies based on these predictions. * While not explicitly stated as future work, the analysis of categorical and numerical features could be expanded to explore more complex interactions or incorporate additional data sources (e.g., user demographics, traffic data).

Summary updated

Abstract

As the sharing economy is booming in China, many intelligent shared parking lots appear. Since more and more Chinese households own cars, it is necessary to study the payment behavior of shared parking lots, which may represent the entire sharing economy. In detail, we analyze the factors that influence users’ payment and predict users’ payment behavior of whether and when users will deliver parking bills after parking. We use 29,733 real parking records provided by Huaching Tech, a top smart parking company in China, in our study. After a comprehensive statistical analysis, we use decision tree model to predict users’ payment behavior. Experiments show that the decision tree model can reach 79% accuracy.
BibTeX
@inproceedings{xu2020payment,
  author = {Qingyu Xu and Feng Zhang and Mingde Zhang and Jidong Zhai and Jiazao Lin and Haidi Liu and Xiaoyong Du},
  title = {{Payment Behavior Prediction and Statistical Analysis for Shared Parking Lots}},
  booktitle = {NPC},
  year = {2020},
  url = {https://openreview.net/forum?id=7KtaHgOYhU},
}
Figure