A handheld laser-induced breakdown spectroscopy dataset of human skeletal remains for forensic reassociation of commingled individuals

Context. When several sets of skeletonized human remains share the same context, individual skeletal elements can become commingled, and reassociating each element with the correct individual is a recurring challenge in forensic anthropology and bioarchaeology, particularly in mass-casualty or disaster-victim-identification settings. Laser-induced breakdown spectroscopy (LIBS) is a rapid, portable and visually non-destructive atomic-emission technique that records the elemental signature of a bone surface, including light elements (H, C, N, O). This dataset was assembled to support the development, evaluation and benchmarking of methods that reassociate commingled human skeletal remains from their LIBS elemental profiles, and to provide an openly available reference of human-bone LIBS spectra for the spectroscopy and forensic communities. Dataset. The dataset contains 1,092 handheld LIBS spectra acquired from six documented adult skeletons (three male, three female; mean age-at-death 71 years, SD 5) from the USF Donated Skeletal Collection (University of South Florida). For each individual, 28 bones distributed across the skeleton were interrogated at a total of 206 anatomical locations (2–14 locations per bone, depending on size and morphology). Spectra were collected with a handheld Z-300 analyzer (SciAps; 1064 nm Nd:YAG laser, 5–6 mJ/pulse) under an internal argon purge; at each location 5 laser cleaning shots removed ~30 µm of surface contamination before 10 accumulated acquisition shots, and each spectrum spans the 190–951 nm spectral range over 23,431 wavelength channels (intensities in arbitrary units). The data are provided as a native MATLAB files (.mat) organized into five matrices: spectra_train (582 × 23,431) and spectra_test (510 × 23,431), their donor labels classes_train (582 × 1) and classes_test (510 × 1), and the common wavelengths axis (1 × 23,431). Rows correspond to spectra and columns to wavelengths; labels encode donor identity (1–6). The training/test partition is bone-disjoint per individual (i.e. the test spectra originate from bones never used for training) so that performance estimates reflect genuine generalization across distinct skeletal elements rather than memorization. Citation. Any reuse of this dataset should cite both this Zenodo deposit and the associated article, "A handheld laser-induced breakdown spectroscopy dataset of human skeletal remains for forensic reassociation of commingled individuals" (publication venue to be confirmed).