A Clovis/Solutrean Projectile-Point Image Classifier

Here I report a Mathematica technique for implementing an artificial-intelligence-based, whole-image classifier of Clovis vs. Solutrean projectile points. The classifier software proper assumes no representation, as such, of empirical properties that are used in traditional archaeology, such as the dimensions of an artifact, flake geometry, the nature of the raw materials, or provenance information. The classifier is trained on 20 images each of Clovis and Solutrean points, and tested on images of 37 Clovis, and 4 Solutrean, points. Given the source classifications of the training and test images, the Wilson score approximation of the 95% confidence interval for the classification of the test items is [0.87, 0.99].