•  
  •  
 

Abstract

Selecting appropriate drug therapy is an important part of clinical decision-making because it requires matching medications with patient complaints and diagnoses. However, drug recommendation in primary healthcare still relies heavily on manual interpretation, limiting scalability, reproducibility, and systematic knowledge extraction from routine medical records. From a Knowledge Engineering and data-driven recommendation perspective, this challenge highlights the need for approaches that can transform clinical co-occurrence patterns into structured, machine-interpretable interactions. This study investigates integrating Association Rule Mining (Apriori and FP-Growth) with Neural Collaborative Filtering (NCF) to develop a context-based drug therapy recommendation system. A total of 1,000 medical record transactions were processed through cleaning, normalisation, and transaction formation, followed by rule mining under four context scenarios. The best configuration was obtained from the combined complaint–diagnosis scenario with min support = 0.01, producing 179 rules and the highest Micro-F1 score of 0.50885. These rules were transformed into weighted context–drug interactions and compared with an NCF model trained on raw interactions. Using Leave-One-Out evaluation with HR@K, NDCG@K, coverage analysis, and significance testing, NCF-RAW outperformed NCF-AR in the full evaluation setting, achieving HR@1 = 0.5556 vs 0.1111, HR@5 = 0.8889 vs 0.5000, and NDCG@5 = 0.7141 vs 0.3027, with significant differences on key metrics. However, no significant differences were found under restricted evaluation with matched coverage. These findings indicate that the effectiveness of combining rule-based mining and deep learning depends strongly on rule coverage and interaction completeness. This study supports more systematic and scalable clinical recommendation modelling from medical record data. Future work may also explore larger datasets and hybrid interaction strategies to improve robustness and clinical validity.

First Page

13

Last Page

25

Download

Share

COinS