Design an Optimized Deep Learning Architecture to Identify Clinical Implications of Adverse Drug Reaction

Abstract

Adverse drug reactions (ADRs) represent considerable public health risks, demanding appropriate monitoring and detection systems. The abundance of user-generated comments on social media platforms provides a viable route for early identification and monitoring of ADRs. This study suggests a hybrid classifier that uses machine learning (ML) and deep learning (DL) approaches combined with social media data to improve ADR detection accuracy. The hybrid classifier model consists of a Recurrent Neural Network (RNN) and a Bi-directional Long-Short Term Method (Bi- LSTM). Social media platforms are a new kind of information source that gives people access to up-to-the-minute data, but they also bring new problems, such as noise and unstructured data. There is a lack of all-inclusive algorithms that can analyze structured and unstructured data to identify ADRs. There is a big obstacle here. Data from several internet message boards is gathered, organized and unstructured data is processed, and keywords are extracted to guarantee correct classification as part of the investigation's multi-pronged strategy. Some feature extraction methodologies include WordNet, TF-IDF, Semantic Similarity, Higher-Order Statistical Features, and Sentiment Analysis. The results validate the usage of the hybrid classifier and show that it is more effective than conventional approaches in ADR recognition, with an accuracy level of 95.6%. A portion of the system enhanced overall performance could be attributed to the Improved Rider Optimization algorithm.