Predicting Drug Risk Level Using Machine Learning Approaches

The study focuses on predicting drug risk levels based on adverse drug reactions (ADRs) using machine learning. With a dataset of nearly a million ADR reports spanning 2011 to 2018 from the Chinese spontaneous reporting database in Jiangsu Province, the researchers addressed the imbalanced nature of the data using Synthetic Minority Oversampling Technique (SMOTE). They proposed a multi-classification framework leveraging SMOTE and various classifiers. Features for classification were derived from ADR signal values calculated using proportional reporting ratio (PRR) or information component (IC). Four classifiers, including Random Forest (RF), Gradient Boost (GB), Logistic Regression (LR), and AdaBoost (ADA), were applied to the data. The optimal combination, PRR-SMOTE-RF, achieved an impressive accuracy rate of 0.95. The study’s findings are expected to provide valuable insights for experts assessing the transition of drugs from prescription to over-the-counter status.