From Similarities to Probabilities: Feature Engineering for Predicting Drugs’ Adverse Reactions

Social media recently became convenient platforms for different groups with common concerns to share their experiences, including Adverse Drug Reactions (ADRs). In this paper, we propose a two stage intelligent algorithm which call “Simi_to_Prob”, that utilizes social forums; ranking ADRs, and evaluating the ADRs prevalence considering age gender as its first stage. second stage, are predicted utilizing data set from Food Administration (FDA). particular, Natural Language Processing (NLP) is used on extract ranked lists of then validated using novel intrinsic evaluation methods. feature engineering extend input space, supervised machine learning method predict future incidences. Our results show correct list three antihypertensive drugs, where high Spearman’s rank correlation coefficients (rs) 0.7458, 0.6678 0.5929 were obtained between SIDER database drug our media. Furthermore, Relatedness achieved area under ROC curve (AUC) reaching 0.959. The showed AUCs 0.96 0.99 prediction probabilities. proposed shows mining can provide reliable source information, additional features be boost methods’ performance in domains Pharmacovigilance research.