Drug Addiction Likelihood Prediction Using a Web-Based Platform

RiskAssess

Abstract

This paper presents RiskAssess as an early feasibility study, a comprehensive web-based platform that integrates machine learning algorithms with clinical workflow to predict drug addiction risk in the Sri Lankan context.. The platform addresses the critical gap between risk identification and intervention by providing immediate counselor assignment, appointment scheduling, and secure messaging capabilities. A multi-step questionnaire covering demographics, environment, relationships, substance use patterns, and mental health was developed and deployed in English, Sinhala, and Tamil. Eight machine learning algorithms were evaluated, with XGBoost, KNN, and ANN achieving the highest accuracy of 90.91%. A hybrid ensemble model combining Random Forest, XGBoost, and SVM was implemented for production, achieving 81.82% accuracy with superior stability and calibrated probability outputs. Feature importance analysis identified motivation to stay sober (17.5%), alcohol use (8.7%), and drug cravings (5.7%) as the most significant predictors. The platform features role-based interfaces for clients, counselors, and administrators, enabling seamless integration from assessment to intervention. Comprehensive security measures including bcrypt password hashing, role-based access control, and audit logging ensure compliance with healthcare data protection standards. The system successfully demonstrates how locally adapted machine learning models can enhance early intervention strategies for substance use disorders while maintaining clinical workflow integration. However, the results are preliminary due to the small dataset size (n=102) and the platform requires further validation before clinical deployment.