Optimizing Adverse Drug Reaction Surveillance: Integrating Latent Semantic Analysis And Artificial Neural Networks In Pharmacovigilance

Abstract

An essential part of pharmacovigilance is ADR (Adverse Drug Reaction) detection, which seeks possible drug side effects. Combining Artificial Neural Networks (ANN) with Latent Semantic Analysis (LSA) may be useful for this purpose. ADR detection identifies adverse drug reactions by examining textual data, such as patient reports, social media posts, or medical records. The goal is to categorize text into "ADR" and "non-ADR." The process of extracting and displaying the contextual meaning of words in a corpus of text is referred to as latent semantic analysis or LSA. The dimensionality of the term-document matrix is decreased by applying Singular Value Decomposition (SVD). For ADR detection, ANN models offer several benefits over conventional classifier algorithms such as Logistic Regression (LR), SVM, and Random Forest (RF); they can integrate multi-modal data, manage intricate non-linear relationships, and comprehend textual context. Therefore, we propose an LSA model with an ANN classifier for detecting ADR from text. Results indicate that the LSA with ANN classifier outperforms traditional classifier algorithms like SVM, LR, and RF etc.