GC-MS-based volatolomics and machine learning for predicting bloodstain time since deposition.

Estimating the time since deposition (TSD) of bloodstains at crime scenes is critical in forensic science. Although various approaches have demonstrated feasibility and accuracy for TSD estimation, analysis of volatile organic compounds (VOCs) offers a promising alternative because it is noninvasive, rapid, and potentially cost-effective. In this study, human blood was chemically profiled using solid-phase microextraction coupled with gas chromatography-mass spectrometry (SPME-GC-MS) to investigate changes in VOC profiles over time and under different storage conditions. A total of 39 metabolites were associated with TSD. This study compared the types, abundances, and temporal patterns of VOCs among three substrates: gauze, wood, and glass. Based on 13 characteristic VOCs commonly detected in dried bloodstains across these three substrates within a 7-day period, a machine learning regression model was developed and evaluated. The optimal model was a random forest (RF) model, achieving root mean square error (RMSE), coefficient of determination (R2), and mean absolute error (MAE) values of 0.8217, 0.9401, and 0.5103, respectively, for the validation set, and 0.7742, 0.8533, and 0.5437, respectively, for the external validation set. These findings demonstrated that volatolomics analysis has strong potential for estimating bloodstain TSD, offering reliable biomarkers, high model accuracy, and robustness to substrate effects. This study provides a new methodological framework for TSD estimation and lays the foundation for further research in this field.