Machine Learning Approach to Predict Drug Consumption Risk Using Personality Traits

Abstract

Background: Worldwide the drug addiction is identified as a major public health problem. The amount of individuals exposed to drug persistently growing. Extreme usage of drug becomes problematic when it origins behavioural disorders, such as addiction. To address these disorderly behaviours, treatment plans should follow a single, standardized, fixed-step approach. The operation to developing scientific knowledge regarding drug abuse has been a distinct part of a public health research agenda for decades. Many potential factors can influence drug use. Through identifying the efficiency of variables the foundation to the drug addiction can be identified. Using machine learning techniques, hidden relationships between variables can be identified. Machine learning has become a convenient tool for classification problems.

Method: This study aimed to explore the impact of the Big Five personality factor impact of drug use associated with individual personality. We use data from the Drug Use Dataset provided by Kaggle. The database contains records for 1885 respondents. For each respondent, 12 attributes were identified as personality measures including NEO-FFI-R. Participants were asked about their use of 18 legal and illegal drugs. For each drug, they had to choose one of the answers: never used the drug, used it more than ten years ago, or in the past decade, year, month, week or day.

Result: The DT algorithm produce best result for the drug Alcohol (94.33). The SVM algorithm provides best results for the drug Cannabis (92.03) and Nicotine (88.84). Among the five factor features the Nscore, Oscore and Cscore provide better contribution to identify the drug consumption risk. The ROC curve (AUC) for all the drugs [Alcohol (0.96), Cannabis (0.98) and Nicotine (0.96)] the SVM Classification Model provide the best accuracy.

Conclusion: From our model it is observed that among the 12 features Age, Neuroticism, Sensation Seeking, Oscore and Cscore contribute higher to predict the drug consumption risk. So, more personality trails are to be conducted by considering the contributing features.