Forecasting Heroin Overdose Occurrences from Crime Incidents

Ertuğrul, Ali Mert
Lin, Yuru
Mair, Christina
Opioid overdoses continue to worsen in the United States, with rapid increases in overdose deaths involving heroin. This crisis, recognized as “opioid epidemic”, has widespread consequences across every region and every demographic group. To enhance the overdose surveillance and to identify the areas in need of prevention effort, in this work, we explore the forecasting capability of heroin overdose occurrences using real-time crime data. Prior works suggested different types of links between the overdose occurrences and criminal activities, such as financial motives and common causes. Grounded on these observations, we present a model that utilizes the spatiotemporal structure of the crime incidents to forecast future heroin overdose occurrences. Results show that, our method achieves better performance, with significantly lower errors (in terms of RMSE and MAE) compared with the baseline method. Our method also allows for meaningful interpretation from both spatial and temporal aspects, including identifying predictive hotspots, local and global contributions, and informative features.
Citation Formats
A. M. Ertuğrul, Y. Lin, and C. Mair, “Forecasting Heroin Overdose Occurrences from Crime Incidents,” 00, 2018, Accessed: 00, 2021. [Online]. Available: