Utilizing Supervised Machine Learning Models for Opioid Hotspot Prediction
Location
Science Center: Bent Corridor
Document Type
Poster - Open Access
Start Date
4-26-2024 12:00 PM
End Date
4-26-2024 2:00 PM
Abstract
The opioid crisis in the USA has been identified as a significant crisis over the past several decades, with documented opioid overdose deaths reaching 400,000 between 1999 and 2018. In particular, Fentanyl, a synthetic opioid 50 times more potent than heroin, contributed to around 31,000 deaths, comprising 65% of all opioid overdose fatalities by 2018 (Marks et al., 2021; 2022). Overdose epidemics have been geographically concentrated, time-specific, and drug-specific, stretching back several years.The current research attempts to train and investigate multiple supervised machine learning models using unrestricted datasets ranging from 2012 to 2021 in order to identify the most effective predictable model for opioid overdose death rates across U.S. counties. We utilized publicly available databases with county-level estimates to derive predictors as fixed effects in our modeling method.The ultimate goal is to use the predictions as a guide tool for strategic decision-making in the investment and allocation of public health services, particularly in hotspot counties, to proactively prevent future risk. Currently, these models demonstrate the remarkable capability to predict rates up to three years in advance, providing a valuable tool for proactive public health interventions. The study is also investigating the social, economic, and demographic features influencing these outcomes, aiming to prioritize and understand their significance. The research findings highlight that among the trained models, the random forest exhibited the best performance given the available data. The results also indicate that population, employee capacity, and annual payroll are some of the most significant socio-economic factors influencing the outcomes. This knowledge enhances our understanding of the complex dynamics associated with the subject matter.
Keywords:
Machine learning, Public health, Opioid crisis
Recommended Citation
Muradi, Aisha and Simoya, Menard, "Utilizing Supervised Machine Learning Models for Opioid Hotspot Prediction" (2024). Research Symposium. 5.
https://digitalcommons.oberlin.edu/researchsymp/2024/posters/5
Major
Computer Science
Award
STRONG Program
Kenneth and Joan Nelson Fund for Student Research
Project Mentor(s)
Adam Eck, Computer Science
2024
Utilizing Supervised Machine Learning Models for Opioid Hotspot Prediction
Science Center: Bent Corridor
The opioid crisis in the USA has been identified as a significant crisis over the past several decades, with documented opioid overdose deaths reaching 400,000 between 1999 and 2018. In particular, Fentanyl, a synthetic opioid 50 times more potent than heroin, contributed to around 31,000 deaths, comprising 65% of all opioid overdose fatalities by 2018 (Marks et al., 2021; 2022). Overdose epidemics have been geographically concentrated, time-specific, and drug-specific, stretching back several years.The current research attempts to train and investigate multiple supervised machine learning models using unrestricted datasets ranging from 2012 to 2021 in order to identify the most effective predictable model for opioid overdose death rates across U.S. counties. We utilized publicly available databases with county-level estimates to derive predictors as fixed effects in our modeling method.The ultimate goal is to use the predictions as a guide tool for strategic decision-making in the investment and allocation of public health services, particularly in hotspot counties, to proactively prevent future risk. Currently, these models demonstrate the remarkable capability to predict rates up to three years in advance, providing a valuable tool for proactive public health interventions. The study is also investigating the social, economic, and demographic features influencing these outcomes, aiming to prioritize and understand their significance. The research findings highlight that among the trained models, the random forest exhibited the best performance given the available data. The results also indicate that population, employee capacity, and annual payroll are some of the most significant socio-economic factors influencing the outcomes. This knowledge enhances our understanding of the complex dynamics associated with the subject matter.
