A Supervised Machine Learning Model for Predicting Adverse Drug Reactions among Tuberculosis Patients in Zimbabwe

Abstract

This dissertation develops a supervised machine learning model to predict adverse drug reactions (ADRs) among tuberculosis (TB) patients in Zimbabwe, addressing a critical need for tools that enhance patient safety and optimize treatment outcomes. The model integrates multimodal data, including chest X-ray images and patient-specific information, to provide comprehensive risk assessments. The study employs the Cross Industry Standard Process for Data Mining (CRISP-DM) framework, using Convolutional Neural Networks (CNNs) to analyze X-rays and ensemble learning techniques to incorporate clinical data. The model achieved high predictive accuracy, with training accuracy reaching 95% and validation accuracy stabilizing at 78%. The confusion matrix analysis demonstrated the model's ability to differentiate between various severity levels of ADRs, facilitating targeted interventions. The research underscores the potential of machine learning in medical diagnostics, particularly in settings with limited resources, and highlights the need for future research to refine the model using larger and more diverse datasets. The findings have significant implications for improving TB treatment and patient outcomes in Zimbabwe.