Yale Machine-Learning Analysis of 11,000 Skeletal Measurements Places Spinosaurus as Fully Aquatic

Researchers measured more than 11,000 skeletal dimensions from hundreds of fossil and extant specimens, then trained simple ML classifiers on living species whose habitats are known. The models incorporated phylogenetic relationships and a WWII-era statistical technique for handling missing data, allowing them to predict soft-tissue swimming adaptations not preserved in bone. Applied to Spinosaurus, the output strongly favored a fully aquatic lifestyle over the previously debated semi-aquatic or terrestrial alternatives.

The finding challenges the long-held view that non-avian dinosaurs remained land-bound. It aligns Spinosaurus with other secondarily aquatic clades that converged on flipper-like limbs and dense bone, suggesting water offered a viable long-term niche even for large theropods. This reframes Cretaceous river ecosystems as containing apex predators that rarely left the water column.

Key limitation is the reliance on skeletal proportions alone; direct evidence of soft-tissue buoyancy control or stomach contents remains absent. Future strengthening would come from three-dimensional biomechanical simulations validated against living crocodilians or from new Moroccan fossils preserving gut contents or skin impressions.