An Ensemble Model for Detection of Adverse Drug Reactions

Abstract

The detection of adverse drug reactions (ADRs) plays a necessary role in comprehending the safety and benefit profiles of medicines. Although spontaneous reporting stays the standard approach for ADR documents, it suffers from significant under reporting rates and limitations in terms of treatment inspection. This study proposes an ensemble model that combines decision trees, support vector machines, random forests, and adaptive boosting (ADA-boost) to improve ADR detection. The experimental evaluation applied the benchmark data set and many preprocessing techniques such as tokenization, stop-word removal, stemming, and utilization of Point-wise Mutual Information. In addition, two term representations, namely, term frequency-inverse document frequency and term frequency, are utilized. The proposed ensemble model achieves an F-measure of 89% on the dataset. The proposed ensemble model shows its ability in detecting ADR to be a favored option in achieving both accuracy and clarity.