TabMixNN: Deep Learning for Mixed-Effects Modeling on Tabular Data

This paper introduces TabMixNN, a PyTorch-based deep learning framework that combines mixed-effects modeling with neural networks for tabular data. It addresses the need for handling hierarchical data and diverse outcome types. The framework's modular architecture, R-style formula interface, DAG constraints, SPDE kernels, and interpretability tools are key innovations. The paper's significance lies in bridging the gap between classical statistical methods and modern deep learning, offering a unified approach for researchers to leverage both interpretability and advanced modeling capabilities. The applications to longitudinal data, genomic prediction, and spatial-temporal modeling highlight its versatility.