Adverse drug events (ADEs) pose a significant threat to patient safety, healthcare costs, and clinical decision-making processes. Traditional ADE detection systems often need to improve reporting and time delays, necessitating exploring advanced machine-learning techniques. This research investigated the potential of deep learning models, particularly BERT-based architectures, for ADE prediction. We developed and fine-tuned BERT-based models for adverse event classification, employing data preprocessing, model training, and performance evaluation techniques. We Normalised the 50000 categories into 20 categories using LDA topic modelling and then trained the model to predict ADEs. The micro-average F1-score (0.89) and accuracy (0.89) indicates the model’s overall performance across all categories, considering each instance equally. Experimental results demonstrated the efficacy of our approach in accurately classifying adverse event reactions across diverse categories. Our findings contribute to advancing pharmacovigilance practices, offering more accurate and reliable ADE prediction systems.