Interpreting Preterm Birth Predictions Using Explainable AI Models

Abstract: Integrating Artificial Intelligence (AI) into healthcare is advancing rapidly, offering transformative potential in healthcare. The paper examines the significance of Responsible AI in healthcare, specifically in the context of preterm birth prediction, with a primary focus on model explainability as a foundational step toward creating informed and trustworthy AI systems. In the context of preterm birth prediction, where existing approaches predominantly rely on image and signal processing techniques, we develop an innovative approach that uses tabular data classification on the CDC Birth dataset. Explainable AI (XAI) is crucial in addressing the ”black box” issue, which hinders trust and understanding between healthcare practitioners and the outputs of algorithms. After classifying preterm birth risk, we employ Local Interpretable Model-agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP) to provide clear, actionable insights into our model’s predictions, thus addressing the ’black box’ problem. By making the model’s decision-making process interpretable, healthcare professionals can better trust AI-driven decisions, enhancing clinical outcomes and ensuring liability. The approach highlights the importance of Responsible AI, emphasising the need for explainability as a cornerstone in deploying AI models in healthcare, particularly for sensitive applications such as predicting preterm birth.