Machine learning-driven research in wood science: from prediction to understanding through the framework of Wood Informatics

Abstract Machine learning (ML) has rapidly expanded across wood science, enabling data-driven approaches for the characterization, performance evaluation, and process optimization of wood and wood-based materials. Although these approaches have demonstrated remarkable predictive accuracy, the field now faces critical challenges related to data heterogeneity, model generalization, and interpretability. Most existing models are developed under narrowly defined experimental conditions, limiting their robustness across different wood species, measurement instruments, and environmental settings. Similarly, while explainable artificial intelligence techniques have enhanced model transparency, their outputs often remain qualitative and insufficiently aligned with the physical and chemical mechanisms governing wood behavior. This review synthesizes the current landscape of ML-driven research in wood science and identifies key challenges for future advancement, emphasizing the need for AI-ready datasets, reliable and generalizable models, and scientifically interpretable approaches. To address these issues, the concept of Wood Informatics is introduced as an integrative framework that connects data standardization, model reliability, and physics-informed interpretability within a unified research ecosystem. By linking prediction to understanding, Wood Informatics—integrating standardized datasets, reliable models, and physically consistent interpretations—establishes a robust foundation for data-centric, reproducible, and explanatory wood science. This transition signifies not only a technological advancement but also a paradigm shift in how wood and wood-based systems are analyzed, understood, and designed.