AIE-DRP: Framework with Machine Learning and Deep Learning Models for Adverse Drug Reaction Prediction in Healthcare Use Case

Abstract

The emergence of Artificial Intelligence (AI) based methods pertaining to machine learning (ML) and deep learning has paved the way for solving many real-world problems. In the healthcare industry, the Quality of Service (QoS) is greatly enhanced with automated approaches in health and diet recommendations, disease prediction and drug recommendations. There is a need for drug reaction prediction as well with AI-enabled approaches. The existing models in this regard suffer from a lack of coordinated Natural Language Processing (NLP) and AI approaches. In this paper, we fill the gap with the proposed framework known as AI-Enabled Drug Reaction Prediction (AIE-DRP). An algorithm known as Multi-Model based ADR Prediction (MM-ADRP). It makes use of multiple ML models such as Logistic Regression, Light Gradient Boosting Machine (LGBM) and MultinomialNB. It also uses deep learning models such as Convolutional Neural Network (CNN) and multiple variants of Long Short Term Memory (LSTM). The framework analyses contextual content and determines whether it has associated adverse drug reaction or not besides categorizing it into either drug or dosage. The experimental results are compared among many models. The CNN model with dropout is found to have the highest prediction performance with 87.35% F1-score. This research can trigger further insights in the healthcare industry to automate adverse drug reaction prediction.