CreditPrint: Credit Investigation via Geographic Footprints by Deep Learning

Xiao Han, Ruiqing Ding, Leye Wang, Hailiang HuangarXiv 2019 · arXiv

Introduction

Traditional credit scoring systems, such as the FICO score, primarily rely on financial histories, including credit card transactions and public records. However, this approach suffers from three significant limitations: richness (excluding "thin-file" individuals like students), timeliness (financial records often lag behind real-life economic changes), and reliability (self-reported data can be easily manipulated).

To address these gaps, the authors propose CreditPrint, a novel deep learning framework that utilizes geographic footprints—GPS trajectories—as a proxy for socio-economic status. The core premise is that our physical movements (where we live, work, and spend leisure time) serve as a continuous, difficult-to-fake indicator of creditworthiness.

The CreditPrint Framework

CreditPrint is a two-stage framework designed to transform raw GPS data into actionable credit scores. It combines spatial understanding with temporal movement patterns.

Overview of the CreditPrint framework, integrating credit-aware region embedding with trajectory-based assessment.

Stage 1: Credit-Aware Region Embedding (REN)

The model first learns the "credit profile" of specific geographic regions (partitioned into 1km x 1km grids). It employs Graph Convolutional Networks (GCNs) to model three types of relationships:

  • Distance Graph: Physical adjacency.
  • Interaction Graph: Frequency of travel between regions.
  • Correlation Graph: Similarity in visitor profiles, regardless of physical distance.

Stage 2: Trajectory-based Credit Assessment (T-CAN)

Once regions are embedded with credit information, the model analyzes individual user paths using Gated Recurrent Units (GRUs) with a Temporal Attention mechanism. This recognizes that certain points in a trip—specifically the destination—are often more indicative of credit than the path taken. The model also incorporates context features like weather and time of week to account for variations in movement patterns.

Key Insights: Locations as Social Filters

A central finding of the research is that geographic regions act as implicit social filters. High-credit individuals tend to cluster in specific professional or high-end residential areas, while low-credit footprints cluster elsewhere. This spatial correlation is a powerful signal for credit investigation.

Spatial distribution maps showing the clustering of default ratios (left) and crime occurrences (right), illustrating how geography reflects socio-economic traits.

In ablation studies, the authors discovered that the Correlation Graph—which links regions with similar visitor credit profiles—was more influential than simple physical distance. Furthermore, the model confirmed that where a person goes (the destination) provides a stronger signal of socio-economic status than the intermediate path.

Experimental Results

The framework was evaluated using a real-life mobility dataset from Hangzhou, China, involving over 5,000 users. Creditworthiness was defined by whether users paid their mobile phone bills on time.

  • Performance: CreditPrint achieved an AUC of 0.784, representing a 10% improvement over traditional manual-feature baselines (which scored between 0.695 and 0.707).
  • Ablation Analysis: Removing the Region Embedding (REN) dropped the AUC to 0.732, while removing the Trajectory Network (TEN) dropped it to 0.723. This demonstrates that both the "where" (location profile) and the "how" (movement patterns) are essential for accurate prediction.

Conclusion and Future Work

CreditPrint demonstrates that geographic footprints are a viable and robust alternative for credit assessment, particularly for individuals lacking traditional financial histories. While the study raises important privacy considerations regarding the use of granular GPS data, the technical results suggest significant potential for geographic data to complement traditional financial metrics. Future work may explore the transferability of these models across different cities and their application in other domains, such as public health monitoring.

Summary updated

Abstract

Credit investigation is critical for financial services. Whereas, traditional methods are often restricted as the employed data hardly provide sufficient, timely and reliable information. With the prevalence of smart mobile devices, peoples' geographic footprints can be automatically and constantly collected nowadays, which provides an unprecedented opportunity for credit investigations. Inspired by the observation that locations are somehow related to peoples' credit level, this research aims to enhance credit investigation with users' geographic footprints. To this end, a two-stage credit investigation framework is designed, namely CreditPrint. In the first stage, CreditPrint explores regions' credit characteristics and learns a credit-aware embedding for each region by considering both each region's individual characteristics and cross-region relationships with graph convolutional networks. In the second stage, a hierarchical attention-based credit assessment network is proposed to aggregate the credit indications from a user's multiple trajectories covering diverse regions. The results on real-life user mobility datasets show that CreditPrint can increase the credit investigation accuracy by up to 10% compared to baseline methods.
BibTeX
@misc{han2019creditprint,
  author = {Xiao Han and Ruiqing Ding and Leye Wang and Hailiang Huang},
  title = {{CreditPrint: Credit Investigation via Geographic Footprints by Deep Learning}},
  year = {2019},
  eprint = {1910.08734},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1910.08734},
}
Figure