Estimating the Temperature of Heat-exposed Bone via Machine Learning Analysis of SCI Color Values: A Pilot Study.

Abstract

Determining maximum heating temperatures of burnt bones is a long-standing problem in forensic science and archaeology. In this pilot study, controlled experiments were used to heat 14 fleshed and defleshed pig vertebrae (wet bones) and archaeological human vertebrae (dry bones) to temperatures of 400, 600, 800, and 1000°C. Specular component included (SCI) color values were recorded from the bone surfaces with a Konica-Minolta cm-2600d spectrophotometer. These color values were regressed onto heating temperature, using both a traditional linear model and the k-nearest neighbor (k-NN) machine-learning algorithm. Mean absolute errors (MAE) were computed for 1000 rounds of temperature prediction. With the k-NN approach, the median MAE prediction errors were 41.6°C for the entire sample, and 20.9°C for the subsample of wet bones. These results indicate that spectrophotometric color measurements combined with machine learning methods can be a viable tool for estimating bone heating temperature.