Technical Note: Using Machine Learning to Predict Locomotor Behavior in Great Apes and Humans From Femur Metaphyseal Shape.

Abstract

<h4>Objectives</h4>The morphology of the hominoid distal femoral metaphyseal surface has been demonstrated to reflect locomotor behavior throughout ontogeny. Here, we quantify metaphyseal surface morphology to evaluate its predictive relationship to locomotor behavioral modes in hominoids.<h4>Materials and methods</h4>We collected three-dimensional (3D) surface laser scans of the femora of 177 human and great ape individuals representing all subadult stages of development. We used the landmark-free Global Point Signature (GPS) method to quantify the shape of the morphologically complex but amorphous metaphyseal surface. We then analyzed the GPS quantifications of shape using support vector machines (SVMs), a machine learning technique, to evaluate the predictive relationships between metaphyseal surface morphology and locomotor behavior in hominoids.<h4>Results</h4>We found that metaphyseal surface morphology is a strong predictor of locomotor behavior in hominoids. Our SVM, which relates nonambulation, bipedal walking, knuckle-walking, and climbing behavior with metaphyseal surface morphology, exhibits ~84% out-of-sample predictive accuracy.<h4>Conclusions</h4>Our quantitative analyses confirm what has previously been qualitatively described-the metaphyseal surface of the distal femur is highly predictive of the locomotor behavior performed by hominoids during different stages of their lives. These results suggest that this region of the skeleton is suitable for reconstructing the locomotor behavior of extinct hominoid taxa.