Nature has evolved highly optimized structural systems that achieve exceptional mechanical performance while minimizing material consumption. Biological architectures such as trabecular bone, honeycomb networks, coral structures, and plant vascular systems have therefore attracted considerable attention as templates for lightweight engineering designs. Despite significant progress in biomimetic structural development, existing studies predominantly focus on individual biological inspirations and evaluate their performance independently. Consequently, the fundamental question of whether universal geometric principles govern mechanical efficiency across diverse natural architectures remains largely unexplored. This study proposes an explainable machine-learning framework to identify and quantify the geometric characteristics responsible for superior mechanical performance in bioinspired lattice structures. A comprehensive biomimetic dataset is constructed by integrating geometric descriptors extracted from multiple nature-inspired architectures, including relative density, connectivity, branching hierarchy, porosity distribution, and fractal complexity. Advanced machine learning algorithms, including Random Forest, Extreme Gradient Boosting, and Artificial Neural Networks, are employed to establish relationships between these descriptors and key mechanical efficiency metrics such as stiffness-to-weight ratio and specific energy absorption. The central novelty of this work is not the development of another predictive machine learning model, but the establishment of a data-driven scientific discovery framework capable of revealing universal design principles embedded within diverse biological systems. Unlike previous studies that treat bone-, honeycomb-, or coral-inspired structures as separate design categories, the proposed approach investigates them collectively to uncover morphology-performance relationships that transcend biological origin. Furthermore, explainable artificial intelligence techniques are integrated to determine the relative importance and interactions of geometric features, enabling direct interpretation of the mechanisms responsible for enhanced mechanical efficiency. This allows the extraction of transferable biomimetic design rules that can be generalized to new structural concepts beyond those represented in the training dataset. The results demonstrate that explainable machine learning can bridge the gap between biological observation and engineering implementation by transforming complex morphological information into actionable design knowledge. The proposed framework provides a scalable pathway for the rapid development of next-generation lightweight structures and establishes a foundation for AI-assisted biomimetic design, where machine learning serves not only as a predictive tool but also as a mechanism for uncovering previously hidden principles of natural structural optimization.