A Machine Learning Approach to Classification of Case Reports on Adverse Drug Reactions Using Text Mining of Expert Opinions

In this paper, we present a machine-learning approach to classify case reports on adverse drug reactions according to the causal relationship of adverse drug reactions (ADR). For this purpose, the Naive Bayes classification algorithm is combined with text mining technique, and applied to textual data of expert opinion on ADR case reports in the Korea Adverse Event Reporting System database. The proposed algorithm classifies the case reports into three categories such as possible, probable, and unlikely based on the causal relationship. Our experimental results show that the precision and recall of data belonging to possible is much higher than the other data belonging to probable and unlikely. We believe that our approach can be used not only for signal but also for prediction and prevention of ADRs.