Maternal health during pregnancy plays a crucial role in forecasting the pregnancy risk level. The objective of this paper is to enhance the maternal risk level prediction by implementing multiple machine learning models and a proposed approach on a tabular dataset that includes Electronic Health Records data and medical historical data, and also by interpreting these models’ predictions using explainable AI. A comparative analysis is conducted on eleven traditional machine learning models and on a proposed stacking framework. These models are evaluated using severe evaluation metrics under a pipeline that combines a nested-cross-validation and Synthetic minority oversampling technique. By incorporating Shapley Additive Explanations value analysis on the most performant models, this analysis interprets the impact of each feature in the prediction. The findings indicate that the proposed stacking model achieved an impressive accuracy and balanced accuracy of 98.7%. Also, based on the Shapley Additive Explanations, the interpretation reveals that medical historical data are more impactful on the prediction than Electronic Health Records data.