Using Machine Learning Techniques to Predict People At-Risk for Drug Addiction: A Bayesian-Based Model

Abstract

Drug addiction is the excessive use of substances such as alcohol, cannabis, cocaine or heroin. It can also take the form of physical or psychological dependence on these substances. The diagnosis of drug addiction is based on a set of behaviors or criteria related to the use of the substance in question. This diagnosis is a complex process that requires questioning and analyzing the behavior of the addict. To this end, mental health practitioners and addictologists predict whether a person is addicted to a particular drug based on many factors, such as the person's environment and family relationships. However, this process is not trivial and requires analysis of previous patient behavior while considering the frequency of substance use. This study proposes a machine learning-based model to measure the risk of substance abuse. The dataset used to develop our predictive models is based on many parameters, such as previous instances of significant addiction in confirmed substance abusers and failures in their lives. A Naïve Bayes machine learning algorithm was used, and the performance of this classifier was measured. The different models developed were evaluated using the most commonly used metrics in machine learning: High Detection Rate, False Alarm, Accuracy, Precision, and F-measure. The results show that using machine learning-based models to predict individuals at risk for drug addiction can greatly assist addiction physicians. Bayesian classification yielded an encouraging accuracy score of 91,4%.