Online E-Health Monitoring and Drug Overdose Prediction Using Machine Learning

Abstract

The opioid crisis is growing out of proportions; this requires smart approaches to detecting patients who are likely to overdose on drugs. This system is using ML to classify heroin consumption using behavioral, psychological and demographic variables that are based on the Drug Consumption Quantified dataset. The twelve critical attributes such as dimensions of personalities and level of impulsivity are used in order to maximize prediction. A normalization of the data is done by StandardScaler followed by classification with Logistic Regression with a balanced class weight, optimized based on GridSearchCV and stratified 5-fold cross-validation. The trained model has a high classification accuracy and reliability with the visual analysis of the accuracy curves and confusion matrices that provides a thorough evaluation of the performance. The framework focuses on interpretability and the importance of mental health indicators in predicting opioid consumption, which ensures the data-grounded and explainable method of overdose risk assessment. The system offers early detection and proactive response to opioid dependency by effective preprocessing, powerful model tuning and visual representation of metrics to develop transparent and trustworthy ML-based analytics.