Authors: Zhiyue Xia*, Department of Geographical Science, University of Maryland, Kathleen Stewart, Department of Geographical Science, University of Maryland, Junchuan Fan, Department of Geographical Science, University of Maryland
Topics: Geography and Urban Health, Medical and Health Geography, Spatial Analysis & Modeling
Keywords: drug activity, drug arrests, substance use, Random Forest, machine learning
Session Type: Paper
Presentation File: No File Uploaded
Substance use disorder affects millions of adolescents and adults in the United States and is a public health concern that impacts every state in the nation. Research on spatial and temporal patterns of drug activities and detecting drug activities clusters is useful for distinguishing hot zones of drug activities, and assessing the availability of substance use treatment facilities in these locations. In this study, we investigated spatial and temporal patterns of more than 37,000 drug arrests in Chicago, IL between 2016 and 2018 where the drug arrest data served as a proxy for where drug activities were located. We designed a Random Forest machine learning model to probe the degree to which certain key locations, built environment factors, and sociodemographic variables were correlated with these patterns, especially with respect to identified hotspot locations for drug arrests in Chicago during this period. The Random Forest classifier was designed to accommodate spatial and temporal autocorrelation in the model learning process and returned the top factors that were estimated to be important for classifying the hotspots as they changed over time, and established significant correlates for drug arrest patterns in general as well as for patterns for particular drug types (e.g., heroin, synthetic drugs) in Chicago.
To access contact information login