The best known and used methods of Chemometrics and Machine Learning are presented here below.

Methods of exploratory analysis
Principal Component Analysis
Principal Component Analysis (PCA) is the major workhorse of the chemometrics tools. PCA method can be used for the following goals:
 Visualisation of X in the multivariate space
 Cluster detection
 Outlier detection
 Effect of variability factors
 Data compression, when reducing X dimensionality
 Noise removal
PCA can be seen as a better way to visualize samples represented by numerous variables, by projecting their original coordinates into a new set of axes, called Principal Components (PC). These axes are built so as to maximize the variance of X, thus extracting the information. The first k components represent the summarized information of X, the last components represent the noise.
Thus, PCA can be seen as a change of axes, designed to better visualize the sample variability, but maintaining the distances and scales between samples. For more convenience, the samples are usually visualized on a 2D ou 3D plane, corresponding to the projection of the samples on this set of 2 or 3 axes
PCA can be the basics for other multivariate methods, such as unsupervised classification or Multivariate Statistical Process Control (MSPC).
Multiblocks Analysis
Multiblock data sets include data sets where
 the same samples are characterized with different blocks of variables, the nature and the number of variables of these different blocks can vary
 or several blocks of samples are characterized with the same variables, the number of samples in each block can be different.
The goal of Multiblock analysis is to identify common and specific information within the different blocks of data.
Independent Component Analysis
ICA aims at identifying the products and phenomena present in a mixture, or during a process. The main components of a PCA, which most often describe a mixture of these pure sources, do not always provide an adequate response.
The basic assumption of the ICA is to consider each row of matrix X as a linear combination of “source” signals, S, with weighting coefficients, or “proportions”, A, proportional to the contribution of the source signals in the corresponding mixtures.
Unlike PCA, the results depend on the number of components you want to extract. Thus, for an ICA with 3 components for example, the first component will be different from that of an ICA performed on 4 components. Some tools exist to make easier the choice of the number of components, such as for example “ICA by block”, which makes it possible to check the robustness of the model by looking at the correlations between the components of ICA models made on data split in blocks.

Unsupervised Classification (clustering) :
For classification purposes, there are two ensembles of methods.
The first one are the “unsupervised classification” methods (or clustering) which aim at regrouping similar samples without the use of prior knowledge. The second one are the “supervised classification” methods (or discrimination), where class memberships are used to build a model.
The clustering techniques are also called “unsupervised classification” techniques in the chemometrics field or “datamining” techniques in the machine learning field. They are exploratory tools which aim at finding “natural” clustering trends out of the inner data structure (X) without any other prior knowledge on the class assignment of the samples. Thus, all these methods measure the similarities between samples only according to their X values.
Hierarchical Clustering Analysis (HCA)
Hierarchical Clustering Analysis is a method which assembles or dissociates successively sets of samples. In Agglomerative Hierarchical Classification, n classes are considered at the beginning, i.e. one class per sample, and are regrouped successively until it constitutes a unique class.
The result is given in a form of a tree, called dendrogram, where the length of the branches represents the distance between groups. The choice of the final groups is decided by cutting at a threshold, defined by the user; thus the number of clusters is not a parameter to be set beforehand.
However, two other criteria must be defined: the distance between samples (generally the Euclidean distance) and the grouping criterion. Different trees are indicated in the figure depending on the grouping criterion chosen. The impact on the classification is quite visible in this figure.
KMeans
Nonhierarchical clustering methods aim at building one final partition of the data. Contrarily to the hierarchical approach, the user must set a fixed number of groups out of his prior knowledge, which can be a strong limitation to these techniques.
The widely used Kmeans method is an iterative procedure used to find the best partition of k classes.
The initial partition of k groups is generally generated at random. Kmeans results are highly dependent on the initial partition and on the choice of the number of classes k.
Then, for each iteration, the barycentre of each class (i.e. the mean vector of the class) is recalculated and the samples are assigned to the nearest centre.
This procedure is carried out until the end criterion is reached (for instance: no assignment changes or maximum number of iterations reached).

Regression methods MLR, PCR, PLS :
MLR – Multiple Linear Regression
Multiple Linear Regression (MLR) is the simplest multivariate modelling method. It is the extent of the simple linear regression to the multivariate case.
This method has the advantage of being straightforward. However, if the explanatory variables are correlated (colinearity), the pseudoinverse matrix calculation leads to unstable models. Another important limitation is that MLR cannot be calibrated if the number of samples is smaller than the number of variables. These two limitations are very often encountered in spectroscopy or imaging where the variables are numerous and strongly collinear. Thus, other modelling methods must be adopted for this kind of data.
PCR – Principal Component Regression
In order to overcome the problems inherent in the MLR method, i.e. the management of collinear data and / or if the number of variables is greater than the number of samples, two steps can be processed:
 The first step consists in applying a PCA on X and extracting from it k informative components (not noisy).
 The components thus extracted can then be used in an MLR instead of X
This method is called Principal Component Regression (PCR). Its disadvantage is that the PCA components are not calculated according to their link with the parameter Y but only according to the maximum variance of X. Y is not always linked to the most important variations in X. Thus, the models are not always very efficient.
PLS – Partial Least Square Regression
Partial Least Squares regression is the most used method in Chemometrics.
Rather than calculating components using only Xvariance as PCA does, the PLS algorithm takes into account the covariance between the variables X and y, i.e. the variances of X and y and the correlation between them.
Components, referred here as latent variables (LV), are then built in order to model y. PLS is thus usually more performing than PCR.
The fewer the number of latent variables, the more robust the model is, i.e. stable with respect to external disturbances, but the more it risks being underfitted. It is therefore necessary to choose the number of latent variables carefully in order to create a performing and robust model.

Supervised discrimination
SIMCA – Soft Independent Modelling of Class Analogy
Soft Independent Modelling of Class Analogy (SIMCA) is based on PCA (Principal Component Analysis) and is thus suitable for high dimensional data.
Each class k is modelled by a specific PCA. This PCA allows to model the intraclass variance. Then, for each model a confidence interval is built to define the membership limit of the class. This limit can be based on the Euclidean distance of the Xresiduals (called Q), on the leverage (or equivalent the Hotelling’s T² or the Mahalanobis distance) or, more often, on the combination of both criteria.
An unknown sample is then classified in class k if it falls within the class limits. A sample can be assigned to several classes if they overlap or are very close to each other, or to none of the classes. In this case, it is possible to consider a “rejection class.
The advantage of SIMCA over other methods of discrimination is that it is very easy to add new classes. Indeed, PCA models are produced by class independently of the other classes.
This method works very well for authenticating products with very different X values. On the other hand, when the signals are very close, methods based on the interclass differences will be preferable.
PLSDA – PLS Discriminant Analysis
PLSDA is a method derived from PLS that allows qualitative analysis or discriminant analysis.
As with PLS, the construction of the PLSDA model is based on the X and y covariance. But unlike PLS, the y of PLSDA are not continuous values. Each column matrix of y corresponds to a class and contains either 1 if the sample belongs to the class, or 0 otherwise (a complete disjunctive coding).
Unlike SIMCA, PLSDA focuses on separating the classes . On the other hand, if the classes are very heterogeneous, this can complicate the modelling because all the samples of the class are assigned to the same quantitative value (1), whereas they are truly different somehow.
The predictions are continuous values because the model still remains a PLS. A new sample will be assigned to the class if its prediction is close to 1 for the associated column. A threshold is usually defined to decide whether or not the sample is assigned to the class, as shown in the figure.

Machine Learning methods
SVM – Support Vector Machines
SVM (Support Vector Machines) method is mostly used for nonlinear or complex problems. It is based on the search for boundaries to separate two classes. Thus, only a part of the calibration samples is actually used: they are the support vectors that delimit the boundaries.
The data are transformed into a new space, called the kernel, which makes it possible to model nonlinearity. In calibration, the dimension of this matrix is nxn (n being the number of samples in calibration). The most common kernel is the Gaussian kernel which requires a Gaussian width optimization parameter: sigma that allows to adjust the degree of nonlinearity. The SVM method also requires the optimization of a regularization parameter which makes it possible to avoid overfitting (C or cost). The adjustment of these two parameters is crucial to obtain a model that is both efficient and robust.
Although originally created for classification, SVM methods have been extended to regression, in particular two methods: SVMR and LSSVM.
ANN – Artificial Neural Networks
Artificial neural networks (ANN), or shallow networks (to differentiate them from deep learning methods), are modelling tools imitating the biological principle of neurons. The most used Artificial Neural Network is the MultiLayer Perceptron (MLP). It is organized in layers of interconnected neurons (fully connected), with, at least 3 layers:
 1 input layer corresponding to the variables X (1 neuron per column)
 1 or more hidden layer (s) of k neurons that correspond to the weights which will have to be trained to build the model
 1 output layer corresponding to the Y (1 neuron per column).
The y can be quantitative values to be predicted or to classes according to the type of network developed. ANN are nonlinear stochastic methods, i.e. each modelling process will lead to a different result. It is generally recommended to carry out several iterations.
Nonlinearities are managed by the use of activation functions at the output of each neuron from the hidden layer(s). Different kinds of activation functions exist (tangent, sigmoid, etc.).
Weights are adjusted by running through each sample of the calibration set several times. A stopping criterion is then necessary to avoid overfitting. These methods are therefore to be used with caution, but modelling tips make it possible to obtain robust models.
CART – Classification And Regression Trees
Classification and Regression Trees (CART) successively split the data into two groups, forming a binary tree.
At each node of the tree, a variable is selected according to its predictive interest, and the optimal separation threshold is calculated. Samples with a value below the threshold are directed to the left of the tree and those greater than or equal to the threshold on the right. Then, each subset is again divided in half from a new variable (which can be the same or different). This process is carried out until all the samples are separated or a minimum per leaf (terminal node) is reached.
In the case of a discrimination model, a new sample will be submitted to the tree and will be assigned to the majority class of the leaf in which it falls after having traversed the tree. In the case of a regression model, the value assigned is the average of the samples from the leaf.
Random Forests
An improvement of this approach, the “Random Forests”, makes it possible to overcome the overfitting problems inherent in the CART method.
The principle is to produce several trees from a bootstrap sampling of the initial data, as well on the samples as on the variables. When a new sample is submitted to the forest, its final prediction corresponds to:
 the average of the predictions of all the trees in the case of a quantitative prediction
 the majority class of all the trees in the case of a classification.
The methods based on the combination of several models are designed as “ensemble methods”, the RF are part of them.
RF allow both to model strong nonlinearities and to deal with the asymmetric distributions of the variables X. They also allow the use of categorical input variables (just like CART) in combination of discrete or continuous quantitative variables.
Several parameters have to be optimized including: the number of trees in the forest, the number of variables to be drawn randomly at each node , and the minimum number of samples per leaf.
Boosting
Boosting methods, like Random Forests, are part of the ensemble methods. On the other hand, unlike RF, boosting models are carried out sequentially.
The most conventional methods use successive shallow CART trees, but it is possible to apply the same principle with other algorithms such as for example SVM. The most famous library is XGBoost, it is composed with several boosting methods.
In the regression case, the LSBoost method can be highlighted. Its principle is to build a first tree to model Y. The second tree is then built in order to predict the residue of Y (to which a part of the initial Y is added in order to achieve a more progressive learning), and so on until obtaining a sufficient number of trees to obtain a successful model.
In the case of discrimination, the AdaBoost algorithm proceeds by weighting the misclassified samples. A first tree makes it possible to carry out a first separation of the classes. A higher weight is then assigned to misclassified samples before building the next tree. The idea is to focus on problematic samples. The process is repeated until a successful model is obtained.
The results of the different trees are combined to obtain the final prediction.

Deep Learning methods
CNN – Convolutional Neural Networks
The Convolutional Neural Networks (CNN) method is the most widely used in the field of deep learning. It is part of the deep neural networks, so there are great similarities with the artificial neuronal networks already mentioned above.
This method was first developed for image recognition. The idea is to add convolutional layers upstream of the “classic” ANN layers. The objective is to automatically extract (by learning) informative features for the targeted objective. For image analysis, these convolutional layers (combined with other parameters: pooling, reLu, etc.) make it possible to overcome the position of the object and its size in the image in order to achieve robust models. Many parameters have to be defined: the number and nature of the layers, the size of the filters, the number of neurons in the hidden layers, etc. It therefore requires a very large number of samples to allow training a very large amount of weights. without overfitting. The complexity of this method, however, makes it a very powerful tool.
This type of method can also be extended to other types of data such as spectroscopic data, by adjusting the parameters appropriately.

Preprocessing
Spectroscopic preprocessing
Spectroscopic data have special features and require a minimum of expertise before they can be used in a model. Spectra can be disturbed by undesirable effects, such as light scattering caused by the structure of the sample, which is the most common perturbation. This is particularly true with near infrared spectra and visible light. Other types of spectroscopy, such as Raman for example, also exhibit undesirable disturbances such as fluorescence, causing strong baselines and thus masking informative peaks.
These kinds of variations can be important if the interest parameter is a physical one (sample structure, particle size, etc.), but they are usually inconvenient for predicting chemical properties.
Some methods to correct additive and multiplicative effects are then applied in order to attenuate these undesirable effects and to help the models to focus on the informative part in the spectrum.
Among the best known preprocessings, there are the baseline corrections (detrend), Standard Normal Variate (SNV), Multiplicative Scatter Correction (MSC), the first and second derivatives (SavitzkyGolay, Norris Gap), smoothing to reduce noise,…
Other advanced methods can also be used such as ExtendedMSC (EMSC), orthogonalization methods (EPO, EROS, DOP, …), …
Dimension reduction
Spectra are continuous signals and therefore collinear. They also often present a very large number of variables. Some methods will therefore not be appropriate, such as MLR or methods for which overfitting can be critical if the number of variables is large, for example ANN (more weights to train). It is thus quite common to apply dimension reduction methods before using these types of models.
The simplest way is to apply a PCA and to extract k components, supposedly informative (not noisy). These components are then used as predictive variables in the chosen method.
A PLS or a PLSDA can also be developed depending on whether one wishes to carry out a quantitative regression model or a discrimination. This allows to extract more informative components than with a PCA.
Variable selection
Variable selection offers several advantages including:
 Principle of parsimony: a less complex model is a more robust model
 Can act as a method of dimension reduction (be careful to the method used later if the extracted wavelengths are correlated)
 Allows you to keep only wavelengths or wavelength ranges that are informative and without noise. This simplifies the model and can make it more efficient thanks to the elimination of the useless wavelengths.
 Can allow choosing a few wavelengths to develop a simpler multispectral instrument.
The most effective method would be to test all possible combinations of variables. Despite advances in the computing power of computers, this is not really possible if the number of variables is large. Strategies must then be used to reduce the number of combinations to be tested. A few methods are given as an example, but many other methods for selecting variables exist.

Stepwise method
The stepwise method can be carried out either forwardly by selecting the informative variables one by one, or backwardly by eliminating the noninformative variables one by one, or by combining both approaches. In the case of forward selection, the principle is to make a model on each of the variables, and the one offering the best performance is selected. In the next step, the combination 2 to 2 of this first variable with all the other variables makes it possible to select a second variable,… one proceeds thus until selecting k variables or until the performances reach a minimum.
In the case of the intervalPLS (iPLS), the selection of contiguous spectral bands is generally preferred rather than isolated wavelengths. Of course, it depends on the objective.

Genetic algortihms
The genetic algorithm method is based on the principle of Darwinian evolution. The idea is to consider that each variable has k genes which are either activated (1) or deactivated (0). To start, the genes are activated randomly, then k models are built on the k subsets of the active variables. The k / 2 less performing subsets of variables are eliminated. For the remaining k / 2, crosses are made by interchanging portions of genes between them (single or double crossover breedings). Each subset is also subject to a probability of random mutation by activating or deactivating certain variables. At the end of this step, again k subsets are obtained and reevaluated. This process is iterated until the stopping criterion is obtained. Throughout the iterations, the most informative variables are identified and kept.
It is possible to consider spectral zones instead of single variables in the case of spectroscopic data.
This method is a stochastic method, so each time the method is launched, a different result can be obtained.

Process supervision methods
MSPC – Multivariate Statistical Process Control
SPC, Statistical Process Control, methods are widely used in order to supervise a process and detect any anomalies or drifts. However, SPC only looks at one parameter at a time, which makes monitoring complicated if the number of parameters to be followed is large; but this method is also less reliable because it does not take into account the interactions between the various variables of the process.
MSPC, Multivariate Statistical Process Control, makes it possible to control the processes in a multivariate way thus taking into account the globality of the process. It is usually based on PCA but other methods can also be used. Leverage and residue statistics are tracked to detect potential issues. Values beyond the statistical confidence intervals are considered as anomalies. It is then possible to return to the variable (s) causing the anomaly in order to diagnose the problem and be able to act quickly on the process.
MSPC can be performed on spectroscopic data (Near infrared, Raman ,…), analytical data (temperature, pressure, concentrations,…), or a combination both.
BSPC – Batch Statistical Process Control
BSPC, Batch Statistical Process Control, is the equivalent of MSPC but for batch process control. The system is therefore more complex.
The most classic method is generally to compare the new batches to a “golden batch”, that represents a reference batch for which the conditions are controlled and optimal. This “golden batch” represents the nominal trajectory. The batches deviating from this trajectory can thus be detected, in the same way as for an MSPC model by following the scores, and / or the residuals and the leverage. It must therefore be well defined from several batches carried out under nominal conditions in order to achieve a robust model with relevant confidence intervals.
Although PCA is the most common method, multiways methods can also be used to follow the process, like for example the PARAFAC method, the data are indeed organized in 3 dimensions: batch x time x variables.
The BSPC faces various challenges including in particular the batch alignment so as to be able to compare them (different duration, key stages occurring at different points, etc.). Several methods make it possible to manage this problem according to the case : synchronization, temporal distortion, normalization, maturity index, etc. This step is critical when the objective is to compare with a “golden standard” and / or when multiways methods are used.

Design of Experiments
Design of Experiments
Using historical databases to model processes requires a very large amount of observations to ensure a minimum of variability. When it is possible, a more rational way consists in choosing the observations or experiments to span the whole desired operating conditions, i.e. the design space, with a maximum of variability. Design of Experiments (DoE) corresponds to that part of chemometrics which aims at planning the relevant experiments, minimizing the cost without decreasing information quality, quantifying the different factor effects, modelling and optimizing the processes.
Different designs correspond to different objectives:
 The DoE methodology often includes a first step which consists in implementing a screening design. The experiments are chosen in order to quantify the influential factors among a large number of factors.
 Full factorial designs are the basic designs which carry out all possible experiments with k twolevel factors, low and high levels. All experiments at the boundaries of the design space are planned.
 Fractional factorial designs are used to screen factors when the number of experiments has to be lowered. The aliase principle allows selection of which experiments from the full factorial design must be running without loosing significant information.
Other screening designs using linear models are also commonly used to identify a few significant factors among many: The PlackettBurman designs, The Rechtschaffner screening designs…
They talk about us
« Ondalys, a bridge between academic research and industry »
Our expertise to make sense of your data
With more than 15 year of experience in data analysis, Chemometrics and Machine Learning, especially applicated to spectroscopic data, our team helps you in each step of your project.
You need a tailormade training ?
Our team study your request to offer you the most suitable and personalized training.