Opioid Abuse Prediction Based on Multi-output Support Vector Regression

Opioid drug abuse has a negative impact on national health and social-economic development. It is essential to provide a solid analysis on the use of drug, efficiently. In this paper, we propose a method for drug use prediction and control. We started with a correlation analysis on historic data on opioid accounting from several states based on K-means cluster- ing. Based on heuristics, we propose our prediction model for opioid accounting based on Multi-output Support Vec- tor Regression (MSVR) while considering population fac- tors. We evaluate our method using drug data in 2017 with several state-of-the-practice baselines. Our proposed MSVR model performs 18% better than the state-of-the-practice ARIMA model on Euclidean loss. Our MSVR model can effectively predict short-term trend of opioid abuse, which can be adopted to opioid abuse prevention.