|
چکیده
|
Drug abuse remains one of the most significant public health challenges globally, affecting millions and resulting in profound social and economic consequences. Traditional detection methods, largely reliant on self-reporting and clinical assessments, often fall short in capturing the complex reality of substance use behaviors. As substance abuse continues to evolve, innovative approaches are required to enhance detection, prevention, and intervention strategies while providing healthcare professionals and policymakers with robust data-driven insights. This study utilized the UCI dataset on drug consumption, comprising 1885 respondents with 12 attributes including personality traits, demographic information, and drug consumption behavior. We employed graph analysis techniques to enhance drug abuse detection, focusing on nicotine consumption. previous work used machine learning, we also used machine learning except that we added new features that we get from the graph. Our methodology involved data preprocessing, correlation analysis using Spearman's coefficient, graph construction with different thresholds for nicotine users and non-users, and feature extraction from the resulting graphs. We extracted seven centrality measures: Degree, Betweenness, Closeness, Eigenvector, Pagerank, Harmonic, and Load Centrality. We then combined these graph-derived features with the original dataset and applied various machine learning models for classification. The results demonstrated strong predictive performance, with the best model (Logistic Regression) achieving an accuracy of 0.985964 and an AUC of 0.999015. Other models, including Histogram-based Gradient Boosting, MLP, and SVM, also showed high accuracy above 0.85. This represents a significant improvement over recent studies in the field of drug abuse detection. Future research should focus on validating these results on diverse external datasets to ensure generalizability. Exploring temporal dynamics within drug abuse networks and integrating advanced techniques like Graph Neural Networks could further enhance the methodology. Additionally, expanding the approach to other substances and behavioral health issues could provide a more comprehensive understanding of addiction patterns. Ethical considerations regarding the use of highly accurate predictive models in healthcare settings should also be carefully addressed to ensure responsible application of these techniques.
|