An analysis of COVID-19's impact on the emergence of bicycle theft in Japan using Gaussian process regression

Abstract

Background

The state of emergency imposed in Tokyo due to COVID-19 led to shifts in urban mobility, potentially altering opportunities for bicycle theft.

Methods

We hypothesized that bicycle theft would decrease during the emergency period and applied Gaussian process regression (GPR) to 34 monthly data points to test this assumption. To address these difficulties, we modeled time-series data on reported bicycle thefts using GPR. We trained a GPR model on data from January 2018 to March 2020 and then applied the trained model to predict theft counts after the state of emergency began in April 2020. The hypothesis was tested by comparing predicted values with actual observed cases.

Results and Conclusions

The results supported the hypothesis. This finding can be explained through routine activity theory, with further analysis showing variation by location. Reductions were greater in single-family homes than in apartments and in bicycle parking lots compared with roads. These results highlight how changes in routine activities, particularly reduced mobility and enhanced protection in certain settings, can shape crime opportunities. The study also demonstrates the value of GPR in analyzing crime trends.