Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Network-Based Transfer Learning Helps Improve Short-Term Crime Prediction Accuracy

Jiahui Wu, Vanessa Frias-Martinez

TL;DR

The paper tackles the problem of short-term, place-based crime prediction when mobility data are scarce in some regions. It proposes a network-based transfer learning framework that transfers weights from data-rich source cities to data-poor target cities to improve predictive accuracy, implemented on a NbConv spatiotemporal model with 8-nearest-neighbor geometry. Empirical results across four US cities and multiple crime types show that majority voting from multiple source cities yields consistent F1 improvements in data-scarce settings, while single-source transfers can hurt performance. Violent crime predictions benefit even more from transfer learning, and smaller target cities experience the largest gains, highlighting the practical potential of cross-city knowledge transfer in resource-constrained environments.

Abstract

Deep learning architectures enhanced with human mobility data have been shown to improve the accuracy of short-term crime prediction models trained with historical crime data. However, human mobility data may be scarce in some regions, negatively impacting the correct training of these models. To address this issue, we propose a novel transfer learning framework for short-term crime prediction models, whereby weights from the deep learning crime prediction models trained in source regions with plenty of mobility data are transferred to target regions to fine-tune their local crime prediction models and improve crime prediction accuracy. Our results show that the proposed transfer learning framework improves the F1 scores for target cities with mobility data scarcity, especially when the number of months of available mobility data is small. We also show that the F1 score improvements are pervasive across different types of crimes and diverse cities in the US.

Network-Based Transfer Learning Helps Improve Short-Term Crime Prediction Accuracy

TL;DR

The paper tackles the problem of short-term, place-based crime prediction when mobility data are scarce in some regions. It proposes a network-based transfer learning framework that transfers weights from data-rich source cities to data-poor target cities to improve predictive accuracy, implemented on a NbConv spatiotemporal model with 8-nearest-neighbor geometry. Empirical results across four US cities and multiple crime types show that majority voting from multiple source cities yields consistent F1 improvements in data-scarce settings, while single-source transfers can hurt performance. Violent crime predictions benefit even more from transfer learning, and smaller target cities experience the largest gains, highlighting the practical potential of cross-city knowledge transfer in resource-constrained environments.

Abstract

Deep learning architectures enhanced with human mobility data have been shown to improve the accuracy of short-term crime prediction models trained with historical crime data. However, human mobility data may be scarce in some regions, negatively impacting the correct training of these models. To address this issue, we propose a novel transfer learning framework for short-term crime prediction models, whereby weights from the deep learning crime prediction models trained in source regions with plenty of mobility data are transferred to target regions to fine-tune their local crime prediction models and improve crime prediction accuracy. Our results show that the proposed transfer learning framework improves the F1 scores for target cities with mobility data scarcity, especially when the number of months of available mobility data is small. We also show that the F1 score improvements are pervasive across different types of crimes and diverse cities in the US.
Paper Structure (14 sections, 3 figures, 7 tables)

This paper contains 14 sections, 3 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Crime Prediction Models and Mobility Data
  4. Transfer Learning
  5. Data
  6. Crime incident data
  7. Human mobility data
  8. Short-term Crime Prediction with Mobility Data
  9. Problem setting
  10. Deep Learning Model
  11. Transfer Learning Framework
  12. Transfer Learning Evaluation Protocol
  13. Results
  14. Conclusion

Figures (3)

  • Figure 1: Arrange the nearest neighbors set for the target census tract $s_1$ and construct the 2D feature map for historical crimes. In the neighboring set of $s_1$, $s_2$ and $s_3$ is the closest to $s_1$; $s_4$ and $s_5$ are the next closest to $s_2$ and $s_3$ respectively; $s_6$ and $s_7$ are the next closest to $s_4$ and $s_5$; $s_8$ and $s_9$ are the next closest to $s_6$ and $s_7$. Similar process is applied to each of the ten mobility features.
  • Figure 2: The framework of the transfer learning technique applied in this study. $Model_s$ and $Model_{s,t}$ have the same network architecture. The parameters $\theta_s$ of the whole architecture of $Model_s$ is transferred to $Model_{s,t}$ as the initialization of $\theta_{s,t}$.
  • Figure 3: Average monthly F1 score for crime prediction by fine-tuned models with transferred knowledge from different source cities.