XAI-Powered MLP Framework for Detecting Adverse Drug Reactions from User Reviews

Abstract

Adverse drug reactions pose significant risks in clinical settings, especially when drug side effects are 
overlooked during early prescription stages. To mitigate such risks, this study focuses on enhancing 
drug side effect prediction using machine learning techniques integrated with Explainable AI (XAI) for 
medical health applications. The core objective is to develop an intelligent, interpretable system that 
not only predicts potential side effects but also provides transparency into the decision-making process, 
fostering trust in healthcare professionals a comprehensive dataset comprising drug attributes, side 
effect profiles, and associated clinical features was used for model training and evaluation. Initial 
experimentation was conducted using various baseline classifiers including Ridge Classifier, Linear 
Support Vector Machine (SVM), Logistic Regression, and Multinomial Naïve Bayes. These models 
served as benchmarks for performance in terms of accuracy, precision, recall, and F1-score. Extensive 
Exploratory Data Analysis (EDA) was performed to uncover patterns, correlations, and imbalances in 
the dataset, aiding in informed feature selection and preprocessing. To improve prediction accuracy and 
enable complex pattern recognition, a Multi-Layer Perceptron (MLP) Classifier was proposed as the 
advanced model. The MLP model, being a deep learning algorithm, demonstrated superior performance 
in capturing nonlinear relationships among features that traditional models often fail to detect. This 
makes the solution viable for real-world deployment in clinical decision support systems (CDSS), 
ensuring safer drug administration and better patient outcomes the project contributes to the field of 
medical AI by delivering a high-performing, interpretable, and reliable solution for drug side effect 
prediction, bridging the gap between complex AI models and practical healthcare applications.