Accurate Age Estimation of Bloodstains Based on Visible Reflectance Spectroscopy and Chemometrics Methods

Abstract

Visible reflectance spectroscopy technique combined with three chemometrics methods were applied to predict the age of bloodstains for forensic purposes. The performance of models established by principal component regression (PCR), partial least-squares regression (PLSR), and least-squares-support vector machines (LS-SVM) methods were compared. Three models, built on three different age periods from 2 to 24 h, one to seven days, and seven to 45 days, were implemented to improve the predictive accuracy. The performance of three LS-SVM models are much better than that of PCR models and just a little better than that of PLSR models. Considering the effect of the specificity of bloodstains on model, LS-SVM model, built on the age period from 2 hours to 45 days, achieved correlation coefficient of the prediction set (Rp) = 0.9900, root mean square error of prediction (RMSEP) = 42.7920 hours, residual predictive deviation (RPD) = 7.6709. Models built on three age periods could significantly improve the predictive capability. The results demonstrate that the visible reflectance spectroscopy combined with LS-SVM would be a reliable tool to accurately estimate the age of bloodstains for forensic practical application.