Predicting potential drug abusers using machine learning techniques

Drug abuse is a noteworthy public health problem in the world. Recently, machine learning has become a favorable tool for classification problem. In this paper, we selected three machine learning algorithms (random forest (RF), Extreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LightGBM) ) to predict potential abuse individuals of methamphetamine and amyl nitrite, two kinds of central stimulants, as well as users’ last consumption time based on personality traits and demographic information. Compared with k-NearestNeighbor, a widely-used classification algorithm, RF, XGBoost and LightGBM have superior performance, with LightGBM being the most efficient one in predicting potential user and estimating usage time. The results of feature importance indicate neuroticism is the most important predictor of drug abuse risk.