Establishment of a random forest regression model to estimate the age of bloodstains based on temporal colorimetric analysis.

ABSTRACT

Bloodstain age estimation is important in forensic science. Although several studies have used spectroscopy to estimate bloodstain ages, this method has not yet been practically applied due to the need for expensive equipment and low reproducibility. Thus, we aimed to develop a bloodstain age estimation model that can be easily performed using a spectrophotometric colorimeter. First, bloodstains were prepared by placing blood obtained from five healthy volunteers on a plastic plate. The bloodstains were kept on conditions with various brightness and temperatures. Then, each bloodstain was dissolved in saline every 24 h to a final concentration of 1%, measured with a spectrophotometric colorimeter, and subjected to machine learning to generate a random forest regression (RFR) model, and finally, the prediction accuracy of the bloodstain age was verified. We also elucidated the mechanism of the color changes utilizing aminoguanidine, which is an inhibitor of Maillard reaction. Finally, we measured the time-dependent color changes of the blood fluids obtained from healthy volunteers and examined if the method could be potentially applied to estimate postmortem interval (PMI). Our results showed that the RFR model estimated the bloodstain age with no substantial assessment, and it was applicable to bloodstains, regardless of the brightness or temperature. The color changes were affected by the addition of aminoguanidine. Furthermore, the method could be applied to blood fluids, suggesting its potential usefulness for PMI estimation. Considering its feasibility, the present method could potentially be introduced to practical forensic sciences in the near future.