Trends in Analytical Chemistry 105 (2018) 191e201

Contents lists available at ScienceDirect

Trends in Analytical Chemistry

journal homepage: www.elsevier.com/locate/trac

Chemometrics in forensic science

Raj Kumar, Vishal Sharma*
Institute of Forensic Science and Criminology, Panjab University, Chandigarh, 160 014, India

a r t i c l e i n f o a b s t r a c t

Article history: This review represents a detailed discussion of the multivariate methods used in the examination of
Available online 23 May 2018 forensic exhibits; their advantages, disadvantages, and efficiency are compared. The last decade has seen
the application of the chemometric methods combined with analytical techniques for characterization
Keywords: and discrimination of samples, which leads to the informative and representative examinations of the
Chemometrics samples. Many research articles with reference to the use of chemometrics in forensic science have been
Forensic science
published. This review has been divided into various sections which include chemometrics, its history,
Analytical techniques
multivariate methods, and the application of chemometrics in various disciplines of forensic science. It is
Discrimination
Characterization
suggested that these new techniques and mathematical/statistical methods should be utilized in forensic
science casework to get statistical confidence in the results.
© 2018 Elsevier B.V. All rights reserved.

1. Introduction The analytical method produces masses of dataset even for a

single sample. For a large number of samples, the amount of output
Forensic science is a discipline in which the examination of data will increase tremendously and it makes the task of expert
complex evidence finds its most critical applications. It is important much tedious, time-consuming and also, the manual examination
to identify the samples correctly and then classify them according can provide false positive results. Therefore, the advanced chemo-
to their class characteristics because the correct identification and metric methods are emerged to analyze the large and complex
classification of each forensic exhibit lead to unbiased verdict in the dataset. Moreover, chemometric methods provide accurate and
legal cases [1]. In this remark, now a day, the scientists have significant results in the quick time domain. There are many che-
increased their interest in the analytical methods such as chro- mometric methods and each method should be used according to
matographic and spectroscopic techniques for the analysis of the type of study: characterization, discrimination, model devel-
complex mixtures such as inks, drugs, paints, glass, etc. opment, etc.
The former tools and techniques might not be appropriate for
today's work and will not solve future's challenges. Here, the con-
Abbreviations: AAS, Atomic Absorption Spectrometry; AFM, Atomic Force Mi- ventional methods refer to oblique UVeVisible light scanning,
croscopy; ANN, Artificial Neural Networks; ATR-FTIR, Attenuated Total Reflectance compound microscopy, color/fluorescence test, a test based on
Fourier Transform Infrared; BLR, Binary Logistic Regression; CA, Cluster Analysis; physical parameters etc. Most of the conventional methods are
CART, Classification and Regression Trees; CNN, Convolutional Neural Network; FID, destructive in nature, less reliable and non-significant whereas, in
Flame Ionization Detector; GA, Genetic Algorithm; GC-MS, Gas Chromatography
Mass Spectrometry; HCA, Hierarchical Cluster Analysis; HPLC, High Pressure Liquid
case of forensic science, advanced, nondestructive and reliable
Chromatography; HSI, Hyperspectral Imaging; ICA, Independent Component methods should be preferred. Therefore, the forensic science lab-
Analysis; ICP-MS, Inductively Coupled Plasma Mass Spectrometry; kNN, k-Nearest oratories must be equipped with advanced techniques and meth-
Neighbor; LA-ICP-MS, Laser Ablation Inductively Coupled Plasma Mass Spectrom- odology to face the forthcoming challenges significantly. In many
etry; LDA, Linear Discriminant Analysis; LIBS, Laser-Induced Breakdown Spectros-
forensic cases, only analytical methods are not sufficient to reach a
copy; LR, Likelihood Ratio; MLR, Multiple Linear Regression Analysis; NAS, National
Academy of Sciences; NIR, Near Infrared; PCA, Principal Component Analysis; PLS- conclusion because the obtained data are very vast and complex.
DA, Partial Least Squares-Discriminant Analysis; RDA, Regularized Discriminant Therefore, the researchers have been adopted chemometric
Analysis; SERS, Surface-Enhanced Raman Spectroscopy; SIMCA, Soft Independent methods to get desirable and significant results. The chemometric
Modeling of Class Analogy; SPA, Successive Projection Algorithm; SW, Stepwise methods give better resolution or separation quality of the samples
Formulation; ToF-SIMS, Time-of-Flight Secondary Ion Mass Spectrometry.
* Corresponding author.
which define its incorporation with analytical (spectroscopic/
E-mail address: vsharma@pu.ac.in (V. Sharma). chromatographic) techniques in recent times [2]. The science of

https://doi.org/10.1016/j.trac.2018.05.010
0165-9936/© 2018 Elsevier B.V. All rights reserved.
192 R. Kumar, V. Sharma / Trends in Analytical Chemistry 105 (2018) 191e201

chemometrics initiates from the chemistry. Many books covering neural networks (ANN); 2nd based on modeling the individual
the topics of chemometrics have been published in the last two classes, i.e., soft independent modeling class analog (SIMCA) [7].
decade [3,4]. However, till date, a little information on the appli-
cability of these techniques to the real casework is available. 3.1.1. Linear discriminant analysis (LDA)
Moreover, these methods are not in practice in the forensic LDA is a technique that builds a mathematical function to
laboratories. maximize the separation between known classes of samples. It
The aim of this review is to discuss the various chemometric condenses the dimensions of the complex dataset by reducing a
tools for the regular analysis and its applicability after fulfilling the huge number of original variables to few new composite di-
pre-requisite criteria. As we know, there is a wide gap between mensions (called as canonical functions), without or with a mini-
forensic analysts and chemometricians, therefore this study will mum loss of information from original [8].
help the forensic expert to provide a detailed description of each LDA explains maximum dissimilarities among pre-defined
chemometric methods along with their utility in forensic science. groups of the sample and the developed model predicts the
group membership of unknown samples. Wilks' l statistic is helpful
2. Chemometrics and associated history for the calculation of the discrimination power of the model and is
defined as;
The exact meaning of Chemometrics is:
detðWÞ
l¼
e the utilization of mathematical and statistical operations in the detðTÞ
field of Sciences (Chemistry) where the outcomes of the
analytical methods are very complex. It extracts the maximum where, det (W) ¼ determinant of with-in group variance-
valuable information from the dataset by using the best mea- covariance matrix; det (T) ¼ determinant of the total var-
surement techniques/ optimal procedures and acquires most of ianceecovariance matrix.
the chemical information from the sample data. It correlates For the significant model, the value of Wilks' l must be smaller.
quality parameters or physical properties of the data. In spite of this method, post hoc classification of the training
dataset can also be used as a method to check the effectiveness of
The term chemometrics was 1st used by scientists Swede, Bruce the discriminant model. The main drawback of LDA is that the
R. Kowalski and Svante Wold in 1972 [2]. In 1974, the establishment matrix inversion is required by the software for the calculation of
of International Chemometrics Society leads to the 1st explanation the matrix. So, this technique is practicable only when the numbers
of the chemometrics. In the year 1980s, some international journals of variables are smaller than the number of samples. This reduction
started special issues regarding the papers on chemometrics. After in data can be achieved by using principal component analysis
that, in 1986e87, two publishers, i.e. Wiley and Elsevier, started (PCA) [2,5].
new chemometrics journals “the Journal of Chemometrics” and
“Chemometrics and Intelligent Laboratory Systems” respectively 3.1.2. Partial least squares discriminant analysis (PLS-DA)
[5]. The problem of the smaller number of variables can be over-
The analytical methods generate a large amount of dataset come by using PLS-DA. This method is based on partial least
posing difficulties in objective interpretation. This issue can squares. It utilizes the independent variables of a matrix ‘X’ and
potentially be overcome through the use of multivariate statistical categorical variables ‘Y’ of the pre-defined sample to develop a
methods. Normally, a chemist is less familiar with such statistical training model, and the group membership of unknown sample can
methods. So, to make the complicated statistical methods practi- be predictive by using the value of the partial least squares of un-
cable for the chemist, the chemometrics is used. For the moment, known samples.
many statistical and numerical software, i.e. SPSS, R, MATLAB, In PLS-DA, a replica of the matrix is formed with ones and zeros.
Minitab, Excel Stat, etc. are available commercially that simplifies This matrix contained columns equal to the classes present and an
this process. observation had the value 1 for the class it belongs to and 0 for the
rest. The original X matrix comprises the preprocessed dataset. The
3. Multivariate data analysis matrices X and Y are disintegrated into the product new two
matrices that contained scores and loadings respectively. This
The spectra/chromatogram obtained through analytical method is different from PCA as PCA uses only the information of
methods vary for different kinds of samples and the chemometric matrix X. However, the PLS method also takes into the account of
methods extract different information to individualize and classify matrix Y. Therefore, the loadings values of X block are calculated
the particular class of samples, this is called as ‘chemical pattern from the scores of the Y block and vice-versa [6,9].
recognition.’ These pattern recognition methods are further divided
into two types i.e. supervised and unsupervised pattern recognition. 3.1.3. k-nearest neighbor (kNN)
kNN is a non-parametric method which is mainly based on the
3.1. Supervised pattern recognitions calculation of the distance between the unknown object to the
training objects. Generally, the method calculates the Euclidean
Supervised pattern recognition methods have been currently distance but one can also use correlation-based measures if the
used for a wide variety of analytical data, with different applications variables are highly correlated with each other. On this basis, the
like individualization, classification, discrimination, fingerprinting unknown sample allocated to the groups that belong to closest of
of samples, and detection of impurities, etc. In these methods, a known group/training group on the basis of majority rule. The k
model is constructed on the basis of samples from the known class. shortest distance between known and unknown samples set is then
This model is further used to predict the class of unknown samples calculated, with a small odd number value of k, the unknown is
[6]. There are two types of approaches for using supervised pattern assigned to the class by kNN methods.
recognition: 1st based on discrimination among the classes, i.e., The method has some advantages over the other supervised
partial least squares discriminant analysis (PLS-DA), linear pattern recognition such as easy to conduct because of simple
discriminant analysis (LDA), k-nearest neighbor (kNN) and artificial mathematical equations, no requirement of normalization, etc.
 R. Kumar, V. Sharma / Trends in Analytical Chemistry 105 (2018) 191e201 193

However, kNN has the similar disadvantage that of LDA and it does space, without correcting the minor errors [12,13]. Other super-
not give good results when the distances among the samples are vised techniques such as UNEQ and classification and regression
large. It doesn't provide a graphical representation of outcome and trees (CART) also exist but there is no literature available for their
hence, has extremely slow computation. application in the field of forensic science.

3.1.4. Soft independent modeling of class analogy (SIMCA) 3.2. Unsupervised pattern recognition
SIMCA is mainly based on the principal component analysis
(PCA). Initially, PCA is utilized to get principal components (PCs) The main unsupervised pattern recognition method is principal
from sample classification, and later on, the best model developed component analysis (PCA). It is the 1st step of data analysis in order
from all kinds of samples is used to the prediction of unknown to detect the patterns in the dataset. PCA primarily reduces the
sample's class. Only those optimal PCs are retained for modeling dimension of the dataset without losing any information from
which explains maximum amount of variation present in the original data. The newly generated few principal components
sample dataset [6]. explain most of the information from the dataset. In addition to PCA
The model provides the best result when the difference between analysis, cluster analysis such as K-mean and hierarchical cluster
the classes is much larger than within-class differences. It also analysis (HCA) is also used as unsupervised pattern recognition
provides results in a graphical manner and hence, the computation method for classification purposes.
is very fast. The plots of loading and scores value obtained after PCA
analysis gives information about the outliers and grouping of 3.2.1. Principal component analysis (PCA)
samples. Moreover, there is the plot for interpretation of SIMCA The main purpose of PCA is to retain most of the variation that is
results which is called as Coomans plot. This plot shows the dif- present in the given data set by reducing the dimensions of a large
ferences between the two classes graphically. The main disadvan- number of interconnected variables to few Principal Components
tage of SIMCA is that it gives poor results when the differences (PCs). From these Principal Components, first few PCs, which are
between the classes are small. uncorrelated with each other, explain the most of variation present
in the dataset. Among all PCs, those PCs are selected whose
3.1.5. Artificial neural networks (ANN) eigenvalue is greater than unity [14]. Thus, the interpretation of
An Artificial neural network (ANN) is a parallel computing de- principal components is easy, but, this simple looking technique
vice, which is basically an attempt to make a computer model of the has vast applications in many disciplines along with a number of
brain. The main objective is to develop a system to perform various different derivations. The linear coefficients of the inverse relation
computational tasks faster than the traditional systems [10]. These of linear combinations are called the component loadings. The
tasks include pattern recognition and classification, approximation, values that represent the samples in the space defined by the
optimization, and data clustering. ANN acquires a large collection of principal components are the component scores. The scores can be
units that are interconnected in some pattern to allow communi- used as input to other multivariate techniques, instead of the
cation between the units. These units also referred to as nodes or original measured variables [15].
neurons, are simple processors which operate in parallel [11]. Each Before analyzing the data by PCA method, there are some as-
neuron has an internal state, which is called an activation signal. sumptions that should be followed to get valid, significant and
Output signals, which are produced by combining the input signals accurate results. However, there is often a solution to overcome this
and activation rule, may be sent to other units. problem if your data violated (not met) certain assumptions. The
A great advantage of ANNs is that causal knowledge of the assumptions are as follow;
relationship between the input and the output variables is not
required. Instead, they learn these relationships through successive a) There should be multiple variables in dataset measured at
training. Moreover, ANNs present remarkable and attractive infor- continuous (i.e. ratio or interval) level. Although, ordinal
mation processing characteristics: (i) non-linearity, allowing better variables can also be used.
fit to the data; (ii) noise insensitivity, providing accurate prediction b) The selected variables should be in a linear relationship
in the presence of uncertain data and measurement errors; (iii) because this method is based on Pearson correlation co-
high parallelism, which implies fast processing and hardware fail- efficients and hence needs to be in a linear relationship.
ure tolerance; and (iv) generalization, enabling application of the c) The sample size should be adequate. The sample size should
model to unlearned data [8]. be large enough to provide a reliable result. There are two
methods which detect sampling adequacy (1): the Kai-
3.1.6. Support vector machine (SVM) sereMeyereOlkin (KMO) and [2] Bartlett's test of Sphericity.
Support vector machine is a statistical method which requires d) The data should be suitable for data reduction. For this,
learning theory of statistics [12]. This method is applicable for the adequate correlations between the variables should be there
classification and regression related problems. For classification, it in order for variables to be reduced to the smaller number of
creates a boundary between the two classes which is independent principal components (PCs).
of the distributions of sample vectors in the dataset. When the e) The data should be free from any significant outliers.
classes are linearly separated, it develops optimal boundaries
which exactly distinguish both the classes and also classify the
unknown samples to representative class. The ‘optimal’ boundary is 3.2.2. Cluster analysis
defined as the most distant hyperplane from both sets, i.e. the It is an exploratory technique used for grouping of samples ac-
‘middle point’ between these sets. The sample vectors which are cording to the types of similarity measures used. The samples
closest to the boundary are called support vectors. having similar signatures of spectra are most likely to get arranged
However, in case of non-linear separation, the boundary is by themselves in the cluster. The number of clusters (fixed) which is
developed by using the kernel method. In this method, a trans- a priority of this technique, is decided by hierarchical cluster
formation of the original vector towards the higher dimensional analysis (HCA). For HCA, Ward's method is used as clustering al-
space takes place, leading to linear separation of the classes. In gorithm and Euclidean distance is used as the similarity measure to
regression method, SVM functions as a linear regression in this explain the scheme of clustering. The best results are obtained with
194 R. Kumar, V. Sharma / Trends in Analytical Chemistry 105 (2018) 191e201

the PCA loading (10 PCs) as an input variable for HCA. The fixed of multidimensional data is generally executed on either raw data
number of clusters is entered to Kemean clustering, which dis- or log-transformed data. However, the data obtained from a spec-
tributes the samples to different clusters according to the method troscopic or chromatographic technique are very complex in na-
employed. This method is used as a combined tool to PCA and for ture. Therefore, these data are first normalized to either mean 0 or
pairwise comparison of the samples. 100 by using various normalization methods. The correct pre-
Further information about exploratory data analysis (EDA) treatment of raw data is an essential part of the valuable and
techniques (e.g., the algorithm of PCA, the definition for the dis- effective analysis. The understanding of basic statistics/mathe-
tance measures and clustering algorithms) can be found in stan- matics is a prerequisite for using such techniques, otherwise
dard chemometric articles and textbooks [2,16]. nonesignificant and misleading results might be obtained.
The use of chemometrics in forensic science is dynamically
increasing from last two decades. Scientists around the world have
4. Application of chemometric methods in the examination
published various research and review on chemometrics, but none
of forensic exhibits
of them has compiled comprehensive information covering up all
the aspects of multivariate analysis in the forensic science disci-
Forensic science is a combination of all the branches of natural
pline. Therefore, the current review amalgamates the different as-
science and allied disciplines, the knowledge of which is essential
pects of multivariate modeling for the investigation of exhibits
and efficient in the dispensation of justice in civil, criminal, and
related to forensic discipline. It also covers usefulness of mathe-
social contexts. The National Academy of Sciences (NAS, USA) [17]
matical information of multivariate models along with their utili-
in their report on the current status of forensic science, have
zation in forensic chemical, physical, biological sciences and
stressed upon the problems being faced by the experts and have
questioned document examination.
also suggested the necessary remedies to overcome these prob-
Some review articles on some specific areas of forensic science
lems. It is emphasized that in order to achieve better results from
have been published; application of serology and gunshot residue
the analysis, the analytical techniques should be computable and
analysis [18], crossing ink lines [19], Application of Surface-
correlated with statistical confidence, with the purpose of mini-
Enhanced Raman Spectroscopy (SERS) [20], atomic force micro-
mizing the errors. Fig. 1 shows different areas of forensic science
scopy (AFM) [21], infrared (IR) spectroscopy and IR imaging [22],
that utilizes the principles of chemometrics.
wildlife forensic [23], analysis of questioned documents [24], bite
In forensic science, the main aim of research is to characterize/
mark identification [25] etc. However, no review is available on
identify a suspected sample or to discriminate/classify a disputed
recent developments and trends of multivariate analysis in forensic
sample to/with its respective groups or to detects the adulteration/
science.
counterfeiting for quality assurance purposes or some other miscel-
laneous applications such as nuclear forensics, soil quality assess-
5. Discussions and prospect
ment, estimation of age/sex, fingerprint enhancement, etc. For the
accomplishment of these aims, various analytical methods i.e. spec-
A survey of literature on the use of multivariate analysis in
troscopy, chromatography, thermal methods, X-Ray based methods
forensic science from 2007 to 2018 is performed using the ISI web of
and microscopic methods are utilized and for the significant result,
knowledge. Most of the articles found which are related to forensic
chemometric methods are employed. The detailed investigation
science and chemometrics. The range of forensic examination by
procedure is represented in Fig. 2 in the form of a flow diagram.
chemometrics methods is certainly wide, including pharmaceutical
Quantitative techniques coupled with statistical software are
tablets, medicines, drugs, cigarette and tobacco, wine, soft drinks,
used to view the similarity or dissimilarity between the sample
paint, textile fibers, gunshot residues, soil, hair dye, bloodstain,
under investigation and standard material used for comparative
semen, bones, saliva, currency note, ballpoint pen ink, cultural
analysis. In the field of forensic science, the statistical interpretation
heritage, toners, etc. Tables 1e4 summarizes the information about
sample type, publication year, the aim of the research and chemo-
metric method utilized in the forensic science disciplines.

Forensic 5.1. Forensic chemical science

Chemical
Sciences The forensic chemical sciences deal with the exhibits such as
Forensic Forensic medicine, soft drinks, wine adulteration, etc. and their identifica-
Biological Physical tion. The spectroscopic methods i.e. Raman Spectroscopy
Sciences Sciences
[26e28,32], Fourier Transform infrared spectroscopy (FTIR) [29,32]
and fluorescence spectroscopy [31], UVeVisible spectroscopy [30],
Forensic Seience are favored in most of the studies because these methods are non-
destructive. Some of the researchers utilize GCeMS [33] and
& UHPLC-MS [34] techniques. The output data are analyzed by
Forensic Chemometrics Forensic
Anthropology Toxicology different multivariate methods to get effective, fast, reliable and
bias results. Some good examples are given as follow;

5.1.1. Y. Roggo et al.

Forensic Y. Roggo et al. [26] utilize Raman spectroscopy coupled with
Forensic
Questioned multivariate analysis for the quality control and detection of
Ballistic
Document counterfeits of pharmaceutical tablets.

Experimental Procedure: The authors have analyzed 25

Fig. 1. Different areas of forensic science which are currently using the principle of different pharmaceutical tablets by using PhAT probe that
natural science along with the utilization of chemometric methods. generates a 6 mm diameter spot which covers the main part of
 R. Kumar, V. Sharma / Trends in Analytical Chemistry 105 (2018) 191e201 195

Fig. 2. Flow chart of investigation procedures involved for the examination of forensic exhibits.

tablets. The scanning is performed in 150e1890 cm1 region tablet. For that, a linear relationship of the area under the curve
with 785 nm laser excitation source. SNV and SavitzkyeGolay and API concentration is observed. The limit of detection ob-
methods are used for data pretreatment. SVM method is used tained is 0.59% API. Secondly, with the help of SVM model, all 25
for identification and classification of the tablets. pharmaceuticals tablets are correctly identified and classified
Findings: The main purpose of the research is to find out the without any error. The developed model has passed the corre-
concentration of API (Active pharmaceutical ingredients) in the lation and API peak test.
196 R. Kumar, V. Sharma / Trends in Analytical Chemistry 105 (2018) 191e201

Table 1
Multivariate analysis of forensic exhibits encountered in forensic chemical science.

Samples types Publication Aims of the research Analytical techniques Applied statistical methods if any Refs.
year

Pharmaceutical tablets 2010 Identification of Pharmaceutical tablets Raman Spectroscopy SVM [26]
Medicine 2011 Counterfeits Detection of Medicine Raman Spectroscopy SVM, PCA, and Correlation [27]
Analysis
Viagra counterfeiting 2011 Counterfeit Viagra Detection Raman microspectrometry imaging PCA, LDA, k-NN, SIMCA [28]
Fireworks 2016 Post-blast residues analysis FTIR PCA [29]
Cigarette tobacco 2011 Discrimination of Cigarette Tobacco UVeVis Spectrophotometry PCA and DA [30]
Wine 2016 Discrimination of Wine Fluorescence Spectroscopy Independent Component [31]
Analysis
Soft drinks 2015 Contamination of Soft Drinks FTIR and Raman spectroscopy Carbon Dots methods [32]
Oil contaminated soils 2013 Removal of Oil contamination from Soils GC-MS CHEMSIC method [33]
Ethanol biomarkers 2017 For identification of chronic alcohol abusers UHPLC-MS/MS PCA and LDA [34]

Table 2
Multivariate analysis of forensic exhibits encountered in forensic physical science.

Samples types Publication Aims of research Analytical techniques Applied statistical Refs.
year methods

Paint 2017 Gamma radiations effects on iodine/epoxy paint FTIR ANOVA-PCA, AComDim method [35]
Textile fibers 2017 Identification and classification of textile fibers ATR-FTIR PCA [36]
Lipstick 2017 Differentiation of red lipsticks ATR-FTIR PCA, CA [37]
Cigarette 2009 Classification of cigarette by successive NIR spectroscopy LDA, SIMCA [38]
projection algorithm
Nuclear forensic 2017 Rapid Nuclear Forensics Analysis via Laser-Based LIBS and laser Raman ANN, PCA [39]
Microphotonic Techniques spectromicroscopy (LRS)
Soil 2015 Challenges in Soil quality assessment Energy dispersive X-ray ANN, HCA, SIMCA, PLS-DA, PCA [40]
fluorescence and scattering
spectroscopy
Glass 2017 Differentiation of Glass samples LA-ICP-MS Likelihood Ratio [41]
Soil 2017 Find the Soil Trace Origin XRD, TGA and Hyper-spectral PCA Analysis [42]
color reflectance (HSI)
GSR 2012 Differentiation of Ammunitions AAS and ICP-AES Regularized discriminant analysis [43]

Table 3
Multivariate analysis of forensic exhibits encountered in forensic biological science.

Samples types Publication Aims of research Analytical techniques Applied statistical Refs.
year methods

Hair dye 2017 Dyed hairs were differentiated by using ATR-FTIR ATR-FTIR PLSDA [44]
Blood stain 2017 Identification and Classification of Bloodstain NIR Spectroscopy SIMCA, PLS-DA, LDA-GA [45]
Semen, Vaginal fluid, 2017 Revealing the location of semen, vaginal fluid, and urine NIR-HSI PCA [46]
and Urine
Semen 2017 Non-destructive screening of Semen samples ATR-FTIR PCA [47]
Body fluids 2016 Identification and differentiation of body fluids Raman Spectroscopy PLS-DA [48]
Brain cancer 2018 Discrimination of human brain tumors from normal structures Raman Spectroscopy and AFM PLS-DA [49]
Human bone 2015 Supervised chemometric methods for sample classification Laser-Induced Breakdown LDA, CART, SIMCA, [50]
Spectroscopy PLS-DA, BLR, NN
Urine 2017 Discrimination of patients with lung cancer and controls Mass spectrometry SVM, PCA-LDA, PLS-DA [51]
Edible Bird's nest 2017 Determine the distribution map of protein contents (PC), Hyper-spectral Imaging Genetic algorithm [52]
carbohydrate contents (CC) and sialic acid Contents (SAC) on PLS (GA-PLS)
Edible bird's nest

5.1.2. P.Y. Sacr

e et al. sample. Further, PCA is able to differentiate between counterfeit
The authors have investigated the Viagra® counterfeiting by and genuine Viagra® on the basis of PC3 because the counterfeit
using chemometrics and Raman microspectroscopic imaging [28]. colored tablets have more intensity. It also confirms the pres-
ence of lactose in the spectral region 830e880 cm1 belongs to
Experimental Procedure: A total of 26 counterfeit and imitation illegal Viagra. The kNN and SIMCA models provide 100% correct
tablets of Viagra® and 08 reference tablets are used in this study. classification of genuine and counterfeit Viagra when analyzed
Each sample is smoothly cut into two parts to avoid spectral the complete spectral range of 200e1800 cm1 whereas LDA
difference caused by the distance between sample and probe. model gives 95.5% of cross-validation results.
The laser wavelength of 785 nm is used throughout the experi-
5.1.3. R. Saad et al.
ments. The data pretreatment is done by MATLAB and SIMCA
R. Saad et al. [31] explain the importance of fluorescence spec-
modeling is performed by using the PLS_toolbox. PCA is used for
troscopy in the discrimination of wine samples by using indepen-
reducing the multidimensional dataset into two dimensional.
dent component analysis (ICA) methods.
Findings: The spectra of genuine and counterfeit Viagra® show
no visible differences in the peak position. However, the in-
Experimental Procedure: In this study, 09 wines belonging to
tensity of counterfeit Viagra® is high in comparison to genuine
three different grape varieties (Shiraz, Cabernet Sauvignon and
 R. Kumar, V. Sharma / Trends in Analytical Chemistry 105 (2018) 191e201 197

Table 4
Multivariate analysis of forensic exhibits encountered in forensic questioned document analysis.

Samples types Publication Aims of research Analytical techniques Applied statistical Refs.
year methods

Currency notes 2013 Discrimination between authentic and Raman Spectroscopy PCA, PLS-DA [53]
counterfeit banknotes
Black ballpoint Inks 2008 Classification and individualization of UVeVis spectroscopy PCA [54]
black ballpoint pen inks
Cultural heritage 2017 Non-invasive characterization of colorants diffuse reflectance infrared Fourier PCA-DA [55]
transform (DRIFT) spectroscopy
Blue ballpoint pen inks 2017 Non-destructive examination of blue ballpoint Diffuse reflectance UVeViseNIR K-mean Cluster, [56]
pen inks in forensic application spectroscopy PCA
Paper 2017 Characterization and discrimination of FTIR PCA [57]
writing/photocopier paper types
Toners 2017 Identify the chemical composition of black Near Infrared Spectroscopy PCA [58]
toners in questioned documents.
Dating of writing inks 2017 Developed two Models, i.e. curve estimation UVeVis Spectroscopy PCA, Multiple Regression [59]
model and MLR model for the dating of blue Analysis
ballpoint pen ink
Fraudulent documents 2017 Hyperspectral infrared images techniques Near and middle infrared PCA and Projection [60]
for fraud detection hyperspectral imaging Pursuit (PP)
Writing/printing papa 2017 Characterization and Discrimination TGA PCA [61]
of paper samples

Pinot Noir) and from 09 different producers are analyzed over a Experimental Procedure: 38 red lipsticks from 20 different
wide range of pH. The opaque nature of sample leads the ex- manufacturers are used to determine the differentiation. All
amination by front face fluorescence spectroscopy because this spectra are collected from 650 to 4000 cm1 region. The
technique examines only the surface of the sample. The spectra cleaning of the crystal is done by 50% isopropanol followed by
are acquired with two excitation wavelengths i.e. 269e490 nm the method to avoid any contamination. PCA, cluster analysis,
and 250e490 nm and two emission wavelengths i.e. and correlation method are used to get significant/desirable
286e540 nm and 286e520 nm. The obtained data are analyzed results.
by ICA method in MATLAB software. Findings: The presence of various oils and waxes are observed
Findings: The fluorescence intensity increases with increase in in all the lipstick samples below 1800 cm1 spectral range. With
the pH because increasing pH leads to the change in the struc- the help of PCA and CA analysis, 9 groups are created: G1eG5
ture of molecules. The discrimination of 45 wine samples alto- and G6 aed. The PCA and CA provide a preliminary classification
gether is achieved by ICA method of 1st 7 ICs. Amongst all plots, of red lipstick as they give 0.29% and 0.51% discriminating power
a scatter plot of IC7 against IC5 provides clear differentiation of respectively and correlation coefficient provides 0.93%
wine samples. The fluorescence signals varied according to the discriminating power and hence, the best discrimination of
pH of wine samples resulting in better discrimination of wine lipstick samples.
samples by ICA method.

The fundamental ideas of these studies that are still present in the
5.2.2. R.S. Corr^
ea et al.
nowadays researches are: (1) Comparison of different algorithms ^a et al. [42] proposed that soils can be differentiated
R.S. Corre
shows that they could all be applied to model similar type of dataset.
via analyzing their clay fraction alone and also, the same soil type
The k-NN and SIMCA give best results in comparison to LDA. How-
can be varied according to the distance they are collected from each
ever, it should be kept in mind that these algorithms are case sensitive
other by using hyperspectral color reflectance and XRD techniques.
and can change in function of the data (2) these methods should be
followed in the routine workflow of forensic science laboratory (3)
Experimental Procedure: 16 each collected soil samples of
the pretreatment of data is an essential step of multivariate analysis,
Oxisol and Inceptisol respectively are air dried and homoge-
otherwise false positive results can be obtained.
nized before the analysis. The organic digestion is done with the
5.2. Forensic physical sciences help of 30% H2O2. The slit and clay fraction are separated by
partial sedimentation process. The clay fraction is examined by
The forensic physical science cases involve paints [35], fibers [36], HSI and XRD methods. The obtained data are submitted for PCA
lipsticks [37], cigarettes [38], nuclear forensic [39], soils [40,42], glass analysis in the origin software.
[41], GSR [43], etc. for the examination of trace and unknown samples Findings: XRD analysis significantly differentiated the soil clay
with the help of analytical methods combined with chemometrics. samples. On the basis of clay components, this technique is able
After the detailed analytical examination, the forensic expert decides, to discriminate Oxisol soil samples over a range of 1 km. The
whether the two items/exhibits are same or not. Experimental pro- color analysis is the most accurate method to group the Incep-
cedure and findings of some of the studies are as follow. tisol samples. This methodology is more effective for discrimi-
nation of Inceptisol samples than Oxisol samples. PCA is able to
5.2.1. M. Gładysz et al. discriminate 97% between Inceptisol and Oxisol samples when
M. Gładysz et al. [37] uses ATR-FTIR spectroscopy combined clay þ XRD and HSI þ clay þ XRD respectively are subjected to
with chemometrics for the identification and differentiation of the analysis. Only XRD and HSI þ XRD in combination with PCA
different types of lipstick brands. provide 100% discrimination.
198 R. Kumar, V. Sharma / Trends in Analytical Chemistry 105 (2018) 191e201

5.2.3. J. Yan~ ez et al. 5.3.2. C.K. Muro et al.

~ ez et al. [43] proposed the conventional as well as a che-
J. Yan C.K. Muro et al. [48] use Raman spectroscopy in combination
mometric method for the differentiation of gunshot residues (GSR) with different classification models for the identification and dif-
on the basis of metallic components present in ammunition brands. ferentiation of various body fluids.

Experimental Procedure: 66 samples from two different Experimental Procedure: A total of 75 samples including pe-
manufactures i.e. FAMAE and CBC are taken from individuals' ripheral blood, semen, saliva, sweat and vaginal fluid are pur-
hands with a cotton swab moistened with 2% EDTA immediately chased. The samples are air dried on microscopic slides. The
after shooting with the firearm. 10% HNO3 is used for dissolution laser of 785 nm is used for spectra collections. The spectral data
of metallic residues. GF-AAS and ICP-OES are used for the are pre-processed in MATLAB software before used for model
detection of metallic components. PCA and Regularized development. PLS_toolbox is used for SVM-DA and PLS-DA
discriminant analysis are used as the multivariate methods. modeling analysis and processing.
Findings: The higher concentrations of Al, Ba, Ca, Cd, Co, Cu, Fe, Findings: All the blood and related products are accurately
K, Hg, Mg, Mo, Pb, Sn and Zn metallic elements in the CBC brand characterized by Raman spectroscopy. Further, the combined
in comparison to FAMAE brands of gunshot residue is deter- classification modeling with different variable selection pro-
mined by AAS and ICP-OES. By the conventional method, two cedures is evaluated and it is found that the chemometric SVM-
brands are differentiated without quantitative certainty. With DA model with GA algorithm variable selection provides 99.9%
the help of regularized discriminant analysis, 100% discrimina- correct classification internally. Moreover, the external cross-
tion and classification of ammunition brands are observed with validation provides 100% correct classification of body fluid
quantitative certainty. samples. The PLS-DA model and SVM-DA model without any
variable selection algorithm provide internal validation errors of
The PCA analysis is a most utilized method in recent time as it 2.3% and 0.9% and external validation errors of 2.1% and 0.5%
reduces the data matrix into few PCs. Some considerations are; (1) respectively.
One should follow all the prerequisites prior to the analysis (2)
cross-validation of obtained results should be done (3) it is better to
apply PCA before any model development to estimate the linear
G. Ramos et al.
5.3.3. A.
relationship among the dataset (4) the advance form of discrimi-
G. Ramos et al. [51] analyzed urine samples of 14 patients
A.
nant analysis may provide better results.
with cancer and 24 control volunteers in order to check the dif-
ferentiation between them.

5.3. Forensic biological sciences

Experimental Procedure: The urine samples from 38 in-
dividuals i.e. 1e24 healthy persons and 25e38 persons having
The forensic biology includes evidence related to hairs, body
lung cancerous are collected. Samples are first treated with NaCl
fluids, bones, urine, etc. Recent literature suggests that chemo-
and then, examined with the help of headspace mass spec-
metrics in combination with analytical methods [44e52] gives
troscopy. Various statistical methods are employed for model
useful information about the discrimination of individuals prior to
development.
DNA analysis. This combined approach provides most promising
Findings: Among different models, PLS-DA provides 91% spec-
results of discrimination/individualization of the suspected person
ificity and 100% sensitivity. The proposed model didn't provide
on the basis of biological evidence as follow;
any false negative results in the screening method. The PCA-LDA
and SVM models provide 100% specificity as well as 100%
sensitivity. The selection of urine samples has advantages over
5.3.1. J.F.Q. Pereira et al.
other fluid such as saliva as it contains more volatiles compo-
J.F.Q. Pereira et al. [45] describe the method of identification of
nents than other biological fluids.
dry blood stains on different surfaces by using NIR spectroscopy
and multivariate analysis.
In most of the aforementioned studies, the authors have
developed multiple models for the comparison of classification of
Experimental Procedure: 31 human blood samples, two animal
their samples and in general, it is a good practice because (1) the
i.e. cat and dog blood samples and different red colored products
accuracy and significance of model depends on the sensitivity of
perceived as blood are collected. NIR spectral ranges of
instruments and the output data (2) parameters like resolution,
908e1676 nm are used for spectra collection. The undesirable
scanning area, scanning speed, etc. used for the analysis of exhibits
effects such as noise, uncontrolled amount of samples etc. are
(3) data pre-treatment methods such as normalization, baseline
corrected by various data pre-processing methods. Different
corrections, noise reduction, etc. This is the reason why one model
modeling methods such as SIMCA, SPA-LDA, GA-LDA, and PLS-
provides better efficiency than others. Therefore, it is suggested
DA are utilized.
that in one type of analysis, the methodology should remain same
Findings: The developed SIMCA model helps in 100% correct
throughout the study.
classification of blood stains deposition on porcelain and glass
substrate but failed in 100% classification of blood stains
deposited on metal and ceramic substrates (80% and 90% 5.4. Forensic questioned document examination
classification respectively). Again, genetic algorithm (GA) vari-
able selection provides better results than successive projec- The questioned documents are examined consistently in
tion algorithm (SPA). GA-LDA model gives 100% correct forensic science laboratories for their authenticity. This includes
classification than SPA-LDA model which provide one false documents associated with checks, question papers, passport du-
positive and one false negative result. PLS-DA model is able to plicity, driving license, or document linked with personal identifi-
discriminate human blood and other stains on all substrates cation such as Aadhar card, voter card, pan card, etc. With their
significantly. widespread use, it is necessary to develop a method through which
 R. Kumar, V. Sharma / Trends in Analytical Chemistry 105 (2018) 191e201 199

the source, individuality, and age of the concerned document can 5.4.3. J.F. Pereira et al.
be established. Some recent studies are as follow; J.F. Pereira et al. [60] evaluate the alteration and deletion in a
questioned document by applying the principle of chemometrics
5.4.1. R. Kumar et al. and HSI in NIR and MID regions.
R. Kumar et al. [57] explored the non-destructive application of
ATR-FTIR technique for characterization and discrimination of pa- Experimental Procedure: Sixteen black ink pen samples are
per samples. used in this study. A straight line is drawn on A4 paper and
analyzed by HSI-MIR and HSI-NIR. Different data pre-processing
Experimental Procedure: Twenty-four types of paper brands methods are used before the analysis. The projection pursuit
are used in this study. The qualitative feature and chemometrics (PP) and PCA methods are used to discriminate the ink samples.
of the obtained spectral data are used for characterization and Findings: It is found that PP shows better results in comparison
discrimination. Three different regions of IR, i.e. 400e2000 cm1, to PCA analysis in HSI-MIR ranges (discrimination of 97.5% and
2000e4000 cm1 and 400e4000 cm1 are selected for differ- 87.5% ink respectively). The HSI-NIR method with PCA and PP
entiation by chemometrics analysis. models describe 76.7% and 83.3% of cross-validation of test ink
Findings: The characterization is achieved by matching the sample respectively while HSI-MIR provides 90% of cross-
peaks with standards of cellulose and inorganic fillers, the usual validation. Therefore, it is concluded that HSI-MIR provides
constituents of paper. It is observed that maximum discrimi- better results in comparison to HSI-NIR.
nation has been procured in the wavenumber range of i.e.
2000e4000 cm1. Again, the discriminating power is better PCA, LDA and regression analysis is used in this discipline of
achieved by chemometrics analysis, PCA i.e. 99.64% rather than forensic science. The conditions of PCA and LDA are the same as
qualitative features i.e. 97.83%. aforementioned. However, there are some prerequisites one should
keep in his mind while estimating the relative age of the document
by the regression model. These are as follows;
5.4.2. V. Sharma and R. Kumar
V. Sharma and R. Kumar [59] proposed a method of the dating of ￮ The storage condition of the document must be same till the
ink by utilizing UVeVis spectroscopy and multiple regression document aged.
analysis approaches. ￮ High humidity conditions enhance the dye degradation.
Therefore, a dry environmental storage should be preferred.
Experimental Procedure: The ink entries are aged for approx. ￮ The initial composition of used ink should match with the dated
nine months (267 Days) and then analyzed by using UVeVis ink.
spectroscopy as shown in Fig. 3. Curve estimation and multiple ￮ High temperature and heating conditions should be avoided for
regression analysis are used to build the models for ink dating. storage of document.
Before developing the model, the factors like the best solvent for
ink extraction and the conditions through which maximum ink Therefore, here it is to be noted that the interpretation of the
is extracted has been determined. The variables are selected by analytical data requires a sound knowledge of prerequisites as well
visual inspection as well as by PCA analysis. as the statistical techniques. These methods are now becoming an
Findings: The cross-validation of curve estimation model pro- essential need for the forensic examiner as well as the basic
vides an error of ±19 days in dating. However, the MLR model chemist, physicist, and biologist working in these areas since they
provides an error of ±10 days in dating which is much signifi- provide accurate and subjective conclusions when applied
cant than curve estimation model. Moreover, all the assump- correctly. Nonetheless, the established statistical protocols provide
tions are met in MLR model significantly. a logical reason that can authenticate the findings of examiners in

Fig. 3. Decrease in absorption of dye w.r.t. increase in time of writing (days).

200 R. Kumar, V. Sharma / Trends in Analytical Chemistry 105 (2018) 191e201

investigation such as identification, differentiation, and classifica-

tion of exhibits. The present review indicates that the fundamental
knowledge of mathematical modeling is still required because
without the basic information of chemometrics, the experts will
not able to get reliable, accurate and significant model for the
predictive data analysis. The current review summarized the detail
description of supervised and unsupervised pattern recognition
methods along with their application in the field of forensic science.
The advantages of automated chemometric methods are dis-
cussed in the present review in the light of amalgamation of these
methods into standard operating procedures. Some special point
should be considered for good results like; sample size should be
large enough to represent variability in the dataset; use of nu-
merical variables in spite of string variables; and the developed
model should be cross-validated. It also suggests that the funda-
mental knowledge of statistics is still needed. Some non-cross-
validated studies would have facilitated so that these studies
Fig. 4. Percentage utilization of chemometric methods on forensic exhibits. could have acquired descriptive and cross-validated results.
For the significant number of samples, the multivariate statis-
tical analysis is expedient due to its ease of interpreting results,
the court of law. The implementation of such methodology also reliability, and speed. It is suggested that one should apply PCA
affords less human errors, less subjectivity, and quick time domain. method before developing any mathematical models as it reduces
Along with these statistical methods, the market survey is also the chances of misclassification many folds. The DA model is the
vital to ascertain the exact source of suspected samples by using a primary choice of authors for the classification purposes. However,
database approach. This type of survey helps in the developing a advanced modeling methods such as SIMCA and SVM gain popu-
forensic database which can further be used for the prediction the larity in recent time. The UNEQ and CART modeling is rarely used in
future unknown samples by using pattern recognition methods. This forensic studies.
type of practices could be used as a complementary method to
classical forensic investigation offering inexpensive, easy, fast and Acknowledgments
non-destructive analysis. Moreover, this strategy fastens the inves-
tigation and hence, helpful in reducing the number of backlogged This work has been financially supported by the Department of
cases in forensic laboratories. The authors feels that still, there is lots Science and Technology (Govt. of India) through sanction no. EMR/
of scope in the research area of chemometric coupled with the 2016/001103 and PURSE-II grant.
analytical methods applied to basic as well as applied sciences.
References
6. Future trends
[1] J.A. Siegel, K. Mirakovits, Forensic Science: the Basics, CRC Press, Boca Raton,
The pattern recognition statistical methods are used for indi- FL, 2010.
[2] M. Otto, Chemometrics: Statistics and Computer Application in Analytical
vidualization and separation of elements from multi-elements ev- Chemistry, Wiley-VCH, Weinheim, 1999.
idence/exhibit. In this review, the maximum research articles [3] P.J. Gemperline, Practical Guide to Chemometrics, second ed., CRC Press, Boca
utilized PCA (36.23%) and discriminant analysis (33.33%) whereas Raton, FL, 2006.
[4] H. Mark, J. Workman, Chemometrics in Spectroscopy, Academic Press,
kNN and others (CHEMSIC Model, Likelihood ratio, etc.) are least Elsevier, 2007.
utilized chemometric methods as shown in Fig. 4. [5] D.L. Massart, B.G.M. Vandeginste, S.N. Deming, Y. Michotte, L. Kaufmann,
The analytical techniques with advantages like fast, reliable, Chemometrics e a Textbook, Elsevier, Amsterdam, 1988.
[6] B.K. Lavine, Encyclopedia of Analytical Chemistry, Wiley, Chichester, 2000.
cost-effective, reproducible and multiple element analysis are [7] R. Todeschini, D. Ballabio, V. Consonni, A. Mauri, M. Pavan, CAIMAN (Classi-
frontier in the analysis of forensic samples, i.e. questioned docu- fication and Influence Matrix Analysis): a new approach to the classification
ment, biological samples, chemical samples and other traces. based on leverage-scaled functions, Chemom. Intell. Lab. Syst. 87 (2007) 3e17.
[8] L.A. Berrueta, R.M. Alonso-Salces, K. Heberger, Supervised pattern recognition
However, the main problem for the analytical technique is to
in food analysis, J. Chromatogr. A 1158 (2007) 196e214.
improve the accuracy of qualitative and quantitative analysis by [9] D. Rogers, A.J. Hopfinger, Application of genetic function approximation to
extracting the useful information from a large number of complex quantitative structure-activity relationships and quantitative structure-
properties relationships, J. Chem. Inf. Comput. Sci. 34 (1994) 854e866.
data i.e. spectra or chromatogram.
[10] I.A. Basheer, M. Hajmeer, Artificial neural networks: fundamentals,
The aforementioned chemometric methods will help the computing, design, and application, J. Microbiol. Methods 43 (2000) 3e31.
forensic experts to correctly identify the exhibits belongs to the [11] E.C. Goncalves, L.A. Minim, J.S.R. Coimbra, V.P.R. Minim, Modeling sterilization
crime and hence the criminals and their modus operandi. In the process of canned foods using artificial neural networks, Chem. Eng. Process.
44 (2005) 1269e1276.
light of the above discussion, some of the above-mentioned ex- [12] C. Cortes, V. Vapnik, Support-vector networks, Mach. Learn. Arch. 20 (1995)
amples explain the usefulness of chemometric methods in forensic 273e297.
science as chemometrics provide robust and accurate results when [13] C.J.C. Burges, A tutorial on support vector machines for pattern recognition,
Data Mining Knowl. Discov. 2 (1998) 121e167.
utilized correctly. It is advisable to follow all the prerequisites of [14] H.F. Kaiser, The application of electronic computers to factor analysis, Educ.
multivariate analysis before applying to real casework otherwise Psychol. Meas. 20 (1960) 141e151.
false positive results might be obtained. [15] J.C. Tewari, J.W.K. Irudayaraj, Floral classification of honey using mid-infrared
spectroscopy and surface acoustic wave based z-Nose Sensor, J. Agric. Food
Chem. 53 (2005) 6955e6966.
7. Conclusions [16] S.F. Møller, J. von Frese, R. Bro, Robust methods for multivariate data analysis,
J. Chemom. 19 (2005) 549e563.
[17] Committee on Identifying the Needs of the Forensic Sciences Community, N.
The incorporation of multivariate methods in forensic analysis is R. C, Strengthening Forensic Science in the United States: a Path Forward,
increasing tremendously as it helps in deciphering all the aspects of National Research Council, Washington, DC, 2009.
 R. Kumar, V. Sharma / Trends in Analytical Chemistry 105 (2018) 191e201 201

[18] K.C. Doty, I.K. Lednev, Raman spectroscopy for forensic purposes: recent ap- [42] R.S. Corre ^a, V.F. Melo, G.G.F. Abreu, M.H. Sousa, J.A. Chaker, J.A. Gomes, Soil
plications for serology and gunshot residue analysis, Trends Anal. Chem. 103 forensics: how far can soil clay analysis distinguish between soil vestiges? Sci.
(2018) 215e222. Justice (2017) 1e7.
[19] L.R. Brito, A.R. Martins, A. Braz, A.B. Chaves, J.W. Braga, M.F. Pimentel, Critical ~ ez, M.P. Farías, V. Zún
[43] J. Yan ~ iga, C. Soto, D. Contreras, E. Pereira, H.D. Mansilla,
review and trends in forensic investigations of crossing ink lines, Trends Anal. R. Saavedra, R. Castillo, P. Sa
ez, Differentiation of two main ammunition
Chem. 94 (2017) 54e69. brands in Chile by Regularized Discriminant Analysis (RDA) of metals in
[20] C. Muehlethaler, M. Leona, J.R. Lombardi, Review of surface-enhanced Raman gunshot residues, Microchem. J. 101 (2012) 43e48.
scattering applications in forensic science, Anal. Chem. 88 (2006) 152e169. [44] M.S. Boll, K.C. Doty, R. Wickenheiser, I.K. Lednev, Differentiation of hair using
[21] G. Pandey, M. Tharmavaram, D. Rawtani, et al., Multifarious applications of ATR FT-IR spectroscopy: a statistical classification of dyed and non-dyed hairs,
atomic force microscopy in forensic science investigations, Forensic Sci. Int. Forensic Chem. 6 (2017) 1e9.
273 (2017) 53e63. [45] J.F.Q. Pereira, C.S. Silva, M.J.L. Vieira, M.F. Pimentel, A. Braz, R.S. Honorato,
[22] A.V. Ewing, S.G. Kazarian, Infrared spectroscopy and spectroscopic imaging in Evaluation and identification of blood stains with handheld NIR spectrometer,
forensic science, Analyst 142 (2017) 257e272. Microchem. J. 133 (2017) 561e566.
[23] R. Ogden, A. Linacre, Wildlife forensic science: a review of genetic geographic [46] F. Zapata, F.E. Ortega-Ojeda, C. García-Ruiz, Revealing the location of semen,
origin assignment, Forensic Sci. Int. Genetics 18 (2015) 152e159. vaginal fluid and urine in stained evidence through near infrared chemical
[24] M. Calcerrada, C. Garcia-Ruiz, Analysis of questioned documents: a review, imaging, Talanta 166 (2017) 292e299.
Anal. Chim. Acta 853 (2015) 143e166. [47] I. Grego
rio, F. Zapata, M. Torre, C. García-Ruiz, Statistical approach for ATR-
[25] M.J. Saks, T. Albright, T.L. Bohan, et al., Forensic bitemark identification: weak FTIR screening of semen in sexual evidence, Talanta 174 (2017) 853e857.
foundations, exaggerated claims, J. Law Biosci. 3 (2016) 538e575. [48] C.K. Muro, K.C. Doty, L.S. Fernandes, I.K. Lednev, Forensic body fluid identifi-
[26] Y. Roggo, K. Degardin, P. Margot, Identification of pharmaceutical tablets by cation and differentiation by Raman spectroscopy, Forensic Chem. 1 (2016)
Raman spectroscopy and chemometrics, Talanta 81 (2010) 988e995. 31e38.
[27] K. De
gardin, Y. Roggo, F. Been, P. Margot, Detection and chemical profiling of [49] H. Abramczyk, A. Imiela, The biochemical, nanomechanical and chemometric
medicine counterfeits by Raman spectroscopy and chemometrics, Anal. Chim. signatures of brain cancer, Spectrochim. Acta Part A: Mol. Biomol. Spectrosc.
Acta 705 (2011) 334e341. 188 (2018) 8e19.
[28] P.Y. Sacre
, E. Deconinck, L. Saerens, T. Beer, P. Courselle, R. Vancauwenberghe, [50] S. Moncayo, S. Manzoor, F. Navarro-Villoslada, J.O. Caceres, Evaluation of su-
P. Chiap, J. Crommen, J.O. Beer, Detection of counterfeit Viagra® by Raman pervised chemometric methods for sample classification by Laser Induced
microspectroscopy imaging and multivariate analysis, J. Pharmaceut. Biomed. Breakdown Spectroscopy, Chemometr. Intell. Lab. Syst. 146 (2015) 354e364.
Anal. 56 (2011) 454e461.

[51] A.G. Ramos, A.P. Anto
n, M.N. S
anchez, J.L.P. Pavo
n, B.M. Cordero, Urinary
[29] C. Martín-Alberca, F. Zapata, H. Carrascosa, F.E. Ortega-Ojeda, C. García-Ruiz, volatile fingerprint based on mass spectrometry for the discrimination of
Study of consumer fireworks post-blast residues by ATR-FTIR, Talanta 149 patients with lung cancer and controls, Talanta 174 (2017) 158e164.
(2016) 257e265. [52] J. Shi, X. Hu, X. Zou, J. Zhao, W. Zhang, M. Holmes, X. Huang, Y. Zhu, Z. Li,
[30] D.L. Giokas, N.C. Thanasoulias, A.G. Vlessidis, Multivariate chemometric T. Shen, X. Zhang, A rapid and nondestructive method to determine the dis-
discrimination of cigarette tobacco blends based on the UVeVis spectrum of tribution map of protein, carbohydrate and sialic acid on Edible bird's nest by
their hydrophilic extracts, J. Hazard Mater. 185 (2011) 86e92. hyper-spectral imaging and chemometrics, Food Chem. 229 (2017) 235e241.
[31] R. Saad, D.J. Bouveresse, N. Locquet, D.N. Rutledge, Using pH variations to [53] M.R. de Almeida, D.N. Correa, W.F.C. Rocha, F.J.O. Scafi, R.J. Poppi, Discrimi-
improve the discrimination of wines by 3D front face fluorescence spectroscopy nation between authentic and counterfeit banknotes using Raman spectros-
associated to Independent Components Analysis, Talanta 153 (2016) 278e284. copy and PLS-DA with uncertainty estimation, Microchem. J. 109 (2013)
[32] P.K. Sarswat, M.L. Free, Light emitting diodes based on carbon dots derived 170e177.
from food, beverage, and combustion wastes, Phys. Chem. Chem. Phys. 17 [54] C.D. Adam, S.L. Sherratt, V.L. Zholobenko, Classification and individualisation
(2015) 27642e27652. of black ballpoint pen inks using principal component analysis of UVeVis
[33] M. Soleimani, M. Farhoudi, J.H. Christensen, Chemometric assessment of absorption spectra, Forensic Sci. Int. 174 (2008) 16e25.
enhanced bioremediation of oil contaminated soils, J. Hazard Mater. 254e255 [55] M. Manfredi, E. Barberis, M. Aceto, E. Marengo, Non-invasive characterization
(2013) 372e381. of colorants by portable diffuse reflectance infrared Fourier transform (DRIFT)
[34] E. Alladio, A. Martyna, A. Salomone, V. Pirro, M. Vincenti, G. Zadora, Evaluation spectroscopy and chemometrics, Spectrochim. Acta Part A: Mol. Biomol.
of direct and indirect ethanol biomarkers using a likelihood ratio approach to Spectrosc. 181 (2017) 171e179.
identify chronic alcohol abusers for forensic purposes, Forensic Sci. Int. 271 [56] R. Kumar, V. Sharma, A novel combined approach of diffuse reflectance
(2017) 13e22. UVeVis-NIR spectroscopy and multivariate analysis for non-destructive ex-
[35] J. Colombani, E. Chauvet, S. Amat, N. Dupuy, D. Gigmes, A FTIR/chemometrics amination of blue ballpoint pen inks in forensic application, Spectrochim. Acta
approach to characterize the gamma radiation effects on iodine/epoxy-paint Part A: Mol. Biomol. Spectrosc. 175 (2017) 67e75.
interactions in Nuclear Power Plants, Anal. Chim. Acta 960 (2017) 53e62. [57] R. Kumar, V. Kumar, V. Sharma, Fourier transform infrared spectroscopy and
[36] P. Peets, I. Leito, J. Pelt, Signe Vahur, Identification and classification of textile chemometrics for the characterization and discrimination of writing/photo-
fibres using ATR-FT-IR spectroscopy with chemometric methods, Spec- copier paper types: application in forensic document examinations, Spec-
trochim. Acta Part A: Mol. Biomol. Spectrosc. 173 (2017) 175e181. trochim. Acta Part A: Mol. Biomol. Spectrosc. 170 (2017) 19e28.
[37] M. Gładysz, M. Kro
l, P. Koscielniak, Differentiation of red lipsticks using the [58] S. Materazzi, R. Risoluti, S. Pinci, F.S. Romolo, New insights in forensic
attenuated total reflection technique supported by two chemometric chemistry: NIR/Chemometrics analysis of toners for questioned documents
methods, Forensic Sci. Int. 280 (2017) 130e138. examination, Talanta 174 (2017) 673e678.
[38] E.D.T. Moreira, M.J.C. Pontes, R.K.H. Galva ~o, M.C.U. Araújoa, Near infrared [59] V. Sharma, R. Kumar, Dating of ballpoint pen writing inks via spectroscopic
reflectance spectrometry classification of cigarettes using the successive and multiple linear regression analysis: a novel approach, Microchem. J. 134
projections algorithm for variable selection, Talanta 79 (2009) 1260e1264. (2017) 104e113.
[39] B. Bhatt, K.H. Angeyo, A.D. Kamadjeu, Rapid nuclear forensics analysis via laser [60] J.F. Pereira, C.S. Silva, A. Braz, M.F. Pimentel, R.S. Honorato, C. Pasquini,
based microphotonic techniques coupled with chemometrics, Energy Proce- P.D. Wentzell, Projection pursuit and PCA associated with near and middle
dia 127 (2017) 76e86. infrared hyperspectral images to investigate forensic cases of fraudulent
[40] M.I. Kaniu, K.H. Angeyo, Challenges in rapid soil quality assessment and op- documents, Microchem. J. 130 (2017) 412e419.
portunities presented by multivariate chemometric energy dispersive X-ray [61] R. Kumar, V. Sharma, N. Verma, P.K. Diwan, V. Kumar, V. Kumar, Analysis of
fluorescence and scattering spectroscopy, Geoderma 241e242 (2015) 32e40. writing/printing paper via thermogravimetric analysis: application in forensic
[41] A. van Es, W. Wiarda, M. Hordijk, I. Alberink, P. Vergeer, Implementation and science, Aust. J. Forensic Sci. (2017) 1e17.
assessment of a likelihood ratio approach for the evaluation of LA-ICP-MS
evidence in forensic glass analysis, Sci. Justice 57 (2017) 181e192.