Estimation of the age of bloodstains from a simulated crime scene using ATR-FTIR spectroscopy and machine learning

Objective: bloodstains serve as objective and stable evidence in criminal proceedings. The bloodstain age provides key information for the investigation and prosecution of crimes, thus bearing significant implications in forensic science. In this study, chromatographic silica gel was used as a bloodstain carrier to simulate the permeable wall surfaces encountered in indoor crime scenes. Bloodstains of different ages were examined for temporal changes using attenuated total reflectance-Fourier transform infrared spectroscopy and machine learning. Methods: venous blood samples were collected from nine healthy volunteers. Fourier transform infrared spectra (4000–600 cm⁻¹) were acquired from each sample at five sampling points over a period of 1–7 days. These spectra were classified using support vector machine, logical regression, random forest, and partial least square discriminant analyses. Subsequently, the spectra were smoothed using a second-order polynomial and a 5-point window. Characteristic bands were selected using the successive projection algorithm and the competitive adaptive reweighted sampling algorithm. Partial least squares regression models were established for the prediction of the bloodstain age with both the full spectra and characteristic bands. Results: the random forest model achieved outstanding classification performance and 99.35% accuracy on the prediction sets. The partial least squares regression model established with second-order smoothing and competitive adaptive reweighted sampling showed the best performance for bloodstain age estimation. For the prediction sets, this model achieved an R_p² of 0.9732, an RMSEP of 0.3335, and an RPD of 6.1065. Conclusion: attenuated total reflectance-Fourier transform infrared spectroscopy can be used for accurate classification of bloodstain samples based on their ages.