Automated Drug Name Mapping Using Weighted Ensemble Natural Language Processing and Machine Learning Reranker

Drug names in Electronic Medical Records (EMR) are often entered by healthcare workers as free text, resulting in spelling errors, inconsistent formatting, and ambiguous abbreviations that hinder automated mapping to standardized drug databases. This study proposes an automated drug name-mapping system that integrates multi-method ensemble scoring with a supervised machine learning reranker. The system employs Term Frequency–Inverse Document Frequency (TF-IDF) character n-grams for candidate retrieval, followed by a weighted ensemble of five similarity methods, including Best Matching 25 and fuzzy string-matching variants, supplemented by pharmaceutical-domain features. A Random Forest reranker is subsequently applied to improve ranking accuracy. Experiments were conducted on 9,326 free-text prescriptions against a master database of 55,078 drug entries, using an 80:20 stratified split to train and evaluate the reranker. The proposed system achieved a Top-1 Accuracy of 79.53%, representing an 18.53 percentage-point improvement over the Euclidean Distance baseline (61.00%). The confidence-scoring mechanism categorized predictions into three tiers: HIGH-confidence cases achieved 87.44% accuracy, and LOW-confidence cases achieved 28.30% accuracy, enabling selective automation for HIGH-confidence cases and targeted manual review for ambiguous cases. These results confirm that ensemble-based Natural Language Processing, combined with machine-learning-based reranking, provides a reliable solution for automated drug name standardization in clinical information systems.