A Synthetic, Clinically Inspired Neonatal Dataset for Multi-Class Disease Prediction and Trust-Aware Machine Learning

This dataset provides a synthetic, clinically inspired collection of neonatal health records designed to support methodological research in machine learning. It focuses on multi-class neonatal disease prediction across nine clinically motivated outcome categories and is intended for studies in trust-aware modeling, uncertainty estimation, confidence-aware decision logic, and calibration analysis.

The dataset is entirely synthetic and does not contain any real patient data. It is designed as a research-enabling resource and is not clinically validated. It must not be used for diagnosis, treatment planning, or real-world medical decision-making.

All features are provided in a single consolidated dataset file. An accompanying metadata file documents clinically inspired feature groupings corresponding to conceptual stages of maternal, birth-related, and postnatal information. These stage definitions are provided for research guidance only and do not impose any constraints on how the dataset must be used.

The dataset has been developed as part of doctoral research focused on advancing trustworthy and transparent machine learning frameworks for neonatal health applications.