Prediction of Drug Risks Consumption by Using Artificial Intelligence Techniques

Abstract

Drug abuse and addiction have reached unprecedented heights, destroying and weakening society. It is considered a dangerous and deadly weapon that has had a significant impact on individuals. Clinical evaluation by experts is the most common method for diagnosing addicted patients and isolating them, but this requires equipment, tools, and human effort. Therefore, in this paper, a new hybridization model (EXT- HBOS) between supervised algorithm (Extra tree) and unsupervised algorithm (histogram-based outlier scores) as well as many states of art machine learning techniques (Extremely Randomized Trees, Cat Boost and Light Gradient Boosting Machine) were used to predict drug-addicted patients based on survey online dataset from Kaggle. The dataset was analyzed, discussed, and rebalanced using random oversampling, also the Grey Wolf Optimization (GWO) algorithm was used for tuning important hyperparameters and get the best one. The results were analyzed and discussed using different performance and statistical methods. The results showed that the hybrid model (EXT- HBOS) did the best on all measures, as well as accuracy and Cohen’s kappa. It gained 90% accuracy score and 74% Cohen’s kappa score. Also, The results illustrated that Neuroticism (Nscore) is the most important factor that tempts an individual to abuse drugs such as heroin.