G. Wang et al.

Microchemical Journal 180 (2022) 107620

Fig. 3. The blood mark data on the napkin carrier were classified according to the PCA analysis score map, PC1 load map, PLSDA analysis score map of human and
non-human classification as well as the evaluation index results of external validation of the model. Fig. 3a: Average spectra of the bloodstains of various species on
napkins carriers. Fig. 3b: The second-derivative average spectra of the bloodstains of various species on napkins carriers. Fig. 3c: PCA analysis score of spectral data of
training set. Fig. 3d: PC1 load diagram of PCA analysis of spectral data of training set. Fig. 3e: PLS-DA analysis classification model. Fig. 3f: PLS-DA analysis clas­
sification model of test set data and model evaluation indicators.

in Fig. 3f, the red dotted line is the boundary. The purple mark above the
boundary indicates human bloodstains, while the yellow mark below the
boundary indicates non-human bloodstains. The prediction results
showed that the classification results of the test data set were 0.767 and
0.963 (sensitivity and specificity), respectively.
Among them, only a small number of spectra of human bloodstains
and two spectra of non-human bloodstains were incorrectly classified,
which also demonstrated the robustness and feasibility of the paper
towel human and non-human bloodstains classification model.
3.3. Identification of cattle and sheep bloodstains on cotton carriers
The non-human bloodstains were separated two samples for cattle
and sheep, PCA analysis results are visible (Fig. 4a), cattle and sheep the
spectral data and chart on PC1 and PC2 explained about 97% of the
variation, cattle (green) general distribution in the middle and lower
side, sheep (blue) on the left side and right side of the upper, sheep
(green) and cattle (blue) cannot be completely separated. Meanwhile, in
the load diagram corresponding to PCA analysis (Fig. 4b), four peaks
played a major role in explaining the difference, among which 1645
cm− 1 and 1531 cm− 1 represented amide I band and amide II band
respectively. It suggested that the difference between cattle and sheep
bloodstains was mainly in the potential change of protein, 1450 cm− 1
indicated the hydrocarbon bending of CH3, and 1390 cm− 1 suggested
that part of the difference might be related to fatty acids and poly­
saccharides. Then, we used PLS-DA algorithm to select 5 LVS to establish
a classification model suitable for cattle and sheep bloodstains on cotton
carrier. In Fig. 4c, the red dotted line indicates whether the sample is
from cattle or sheep, the green square above the dividing line indicates
that the sample is from cattle, and the blue triangle below the dividing
line indicates that the sample is from sheep. The classification indexes of
cattle and sheep bloodstains were 0.840 and 0.793 (sensitivity and
specificity), respectively. The test data set was loaded into the PLS-DA
model, and the results are shown in Fig. 4d. The red dotted line is the
boundary, the green square above the boundary represents cattle’s
blood, and the blue triangle below the boundary represents sheep’s
blood. The prediction results showed that the classification results of the
test data set were 0.792 and 0.633 (sensitivity and specificity),