Enhancing forensic investigations: Identifying bloodstains on various substrates through ATR-FTIR spectroscopy combined with machine learning algorithms
Chun-Ta Wei,Jhu-Lin You,Shiuh-Ku Weng,Shun-Yi Jian,Jeff Cheng-Lung Lee,Tang-Lun Chiang
DOI: https://doi.org/10.1016/j.saa.2023.123755
2024-03-05
Abstract:The forensic analysis of bloodstains on various substrates plays a crucial role in criminal investigations. This study presents a novel approach for analyzing bloodstains using Attenuated Total Reflectance Fourier Transform Infrared spectroscopy (ATR-FTIR) in combination with machine learning. ATR-FTIR offers non-destructive and non-invasive advantages, requiring minimal sample preparation. By detecting specific chemical bonds in blood components, it enables the differentiation of various body fluids. However, the subjective interpretation of the spectra poses challenges in distinguishing different fluids. To address this, we employ machine learning techniques. Machine learning is extensively used in chemometrics to analyze chemical data, build models, and extract useful information. This includes both unsupervised learning and supervised learning methods, which provide objective characterization and differentiation. The focus of this study was to identify human and porcine blood on substrates using ATR-FTIR spectroscopy. The substrates included paper, plastic, cloth, and wood. Data preprocessing was performed using Principal Component Analysis (PCA) to reduce dimensionality and analyze latent variables. Subsequently, six machine learning algorithms were used to build classification models and compare their performance. These algorithms comprise Partial Least Squares Discriminant Analysis (PLS-DA), Decision Trees (DT), Logistic Regression (LR), Naive Bayes Classifier (NBC), Support Vector Machine (SVM), and Neural Network (NN). The results indicate that the PCA-NN model provides the optimal solution on most substrates. Although ATR-FTIR spectroscopy combined with machine learning effectively identifies bloodstains on substrates, the performance of different identification models still varies based on the type of substrate. The integration of these disciplines enables researchers to harness the power of data-driven approaches for solving complex forensic problems. The objective differentiation of bloodstains using machine learning holds significant implications for criminal investigations. This technique offers a non-destructive, simple, selective, and rapid approach for forensic analysis, thereby assisting forensic scientists and investigators in determining crucial evidence related to bloodstains.