Anomaly Detection in Spatiotemporal Patterns: A Case Study of Brazilian Traffic Accidents During the COVID-19 Lockdown

• Gabriel César Silveira UFSC
• Ricardo Jose Pfitscher UFSC
• Benjamin Grando Moreira UFSC
• Diogo Nardelli Siebert UFSC

Resumo

This study proposes a methodology to assess pattern changes in spatio-temporal data as a descriptive analysis. As a study case, we focus on traffic accident profile changes in Brazil due to the lockdown of the COVID-19 pandemic, analyzing different categories of occurrences separately. The method uses mainly public data on highway incidents and applies spatial correlation and linear regression models. The combined indicator of dissimilarity and spatiotemporal autocorrelation identifies regional anomalies when regressed regarding data before and during the pandemic. The methodology revealed that the lockdown affected accident characteristics differently across states, leading to changes in state-level and regional rates that could be overlooked in exploratory analyses or when neglecting spatial relationships. This case study can inform public policies and guide future research on the impacts of lockdown traffic and on human behavior.