Physicochemical Problems

Physicochem. Probl. Miner. Process. 50(2), 2014, 575−589

of Mineral Processing
ISSN 1643-1049 (print)
www.minproc.pwr.wroc.pl/journal/
ISSN 2084-4735 (online)

Received November 4, 2013; reviewed; accepted January 11, 2014

MULTI-PARAMETER DATA VISUALIZATION BY MEANS

OF PRINCIPAL COMPONENT ANALYSIS (PCA) IN
QUALITATIVE EVALUATION OF VARIOUS COAL TYPES

Tomasz NIEDOBA

AGH University of Science and Technology, Faculty of Mining and Geoengineering, Department of
Environmental Engineering and Mineral Processing, al. Mickiewicza 30, 30-059 Krakow, [email protected]

Abstract: Multi-parameter data visualization methods are a modern tool allowing to classify some
analyzed objects. When it comes to grained materials, e.g. coal, many characteristics have an influence on
the material quality. Besides the most obvious features like particle size, particle density or ash contents,
coal has many other qualities which show significant differences between the studied types of material.
The paper presents the possibility of applying visualization techniques for coal type identification and
determination of significant differences between various types of coal. The Principal Component Analysis
was applied to achieve this purpose. Three types of coal 31, 34.2 and 35 (according to Polish
classification of coal types) were investigated, which were initially screened on sieves and subsequently
divided into density fractions. Next, each size-density fraction was analyzed chemically to obtain other
characteristics. It was pointed out that the applied methodology allowed to identify certain coal types
efficiently, which makes it useful as a qualitative criterion for grained materials. However, it was
impossible to provide such identification based on contrastive comparisons of all three types of coal. The
presented methodology is a new way of analyzing data concerning widely understood mineral processing.

Keywords: Principal Component Analysis, PCA, multi-parameter data visualization, coal, identification
of data, covariance matrix, pattern recognition

Introduction
In mineral processing operations particle properties influence processing results
(Drzymala, 2009; Kelly and Spottiswood, 1989). In the case of coal the most
important parameters are size and density. But it is often not enough to classify coal
type properly. In modern statistical research there is often a need of applying multi-
parameter (called also as multidimensional) statistical methods. A simple regressive
analysis is not enough in more complex cases (Brozek and Surowiak, 2005; 2007,
2010). That is why many new techniques are introduced in scientific works. Of course,

http://dx.doi.org/10.5277/ppmp140213
576 T. Niedoba

there are many methods connected with regressive equations (Gawenda et al., 2005;
Lyman, 1993; Niedoba, 2009; 2011; 2013a; 2013b; Niedoba and Surowiak, 2012;
Saramak, 2011; 2013; Snopkowski and Napieraj, 2012; Tumidajski, 1997; Tumidajski
and Saramak, 2009) or even fractals (Ahmed and Drzymala, 2005) but many methods
of data-mining are also in use. A special type of statistical analysis involves multi-
parameter data visualization methods geared towards recognizing differences and
similarities between analyzed sets of data. Finding these differences is often a very
important issue in mineral processing, where processes depend on many material
features.
The qualitative analysis of multi-parameter data (properties of material) obtained
from the results of empirical experiments can be carried out by applying the multi-
parameter visualization method. The results of analyses can be helpful thanks to the
characteristics of materials as well as the development of mineral processing models
based on this data. Attempts to depict multi-parameter data have been undertaken on
many occasions. Among many methods, the following ones can be selected: grand-
tour method (Asimov, 1985, Cook et al., 1995), use of neural networks for data
visualization (Aldrich, 1998; Jain and Mao, 1992; Kraaijveld et al., 1995), parallel
coordinates method (Chatterjee et al., 1993; Chou et al., 1999; Gennings et al., 1999;
Inselberg, 1985), star graph method (Sobol and Klein, 1989), multidimensional scaling
(Kim et al., 2000), scatter-plot matrices method (Cleveland, 1984), relevance maps
method (Assa et al., 1999). Visualization of multidimensional solids is also possible
(Jamroz, 2001; 2009). The observational tunnels method (Jamroz, 2001; 2014) makes
it possible to obtain an external view of the observed multi-parameter sets of points
using tunnel radius introduced by the present author (Jamroz and Niedoba, 2014;
Niedoba and Jamroz, 2013).
The use of methods of multi-parameter data visualization by transformation of
multidimensional space into two-dimensional space makes it possible to show multi-
parameter data on computer screen. This allows to conduct a qualitative data analysis
in the most natural way for human with a sense of sight. One of such methods is the
Principal Component Analysis (PCA). It was used in the work to present and analyze
a set of seven-parameter data describing samples of three various coal types 31, 34.2
and 35 (according to Polish classification of coals). It was decided to check whether
this method allows to state that an amount of information contained in seven coal
features is sufficient for the proper classification of coal types. The application of
various methods for analyzing possibilities of recognition of various coal properties is
becoming an increasingly interesting issue. Earlier, other visualization methods were
applied, including the observational tunnels method (Jamroz and Niedoba, 2013;
2014). An application of the PCA method for this purpose is a new way of
approaching the subject.
 Multi-parameter data visualization by means of principal component analysis… 577

Material characteristics
Three types of coal, types 31 (energetic coal), 34.2 (semi-coking coal) and 35 (coking
coal) in the Polish classification were used in the investigation (Olejnik et al., 2010).
The classification of coal types can be found in (Drzymala, 2009). The coals
originated from three various Polish mines and all of them were initially screened on a
set of sieves of the following sizes: –1.00, –3.15, –6.30, –8.00, –10.00, –12,50,
–14.00, –16.00 and –20.00 mm. Subsequently, the size fractions were additionally
separated into density fractions by separation in dense media using zinc chloride
aqueous solution of various densities (1.3, 1.4, 1.5, 1.6, 1.7, 1.8 and 1.9 g/cm3). The
fractions were used as a framework for further consideration and additional coal
features were determined by means of chemical analysis. For each density-size
fraction the following parameters were determined: combustion heat, ash contents,
sulfur contents, volatile parts contents and analytical moisture, making up, together
with the mass of these fractions, seven various features for each coal. The examples of
such data were presented in Table 1 containing the data for size fractions of 14.00-
12.50 mm for each type of coal. The complete data can be found in Niedoba (2013a).

Principal Component Analysis

Method description
The PCA method is one of statistical methods of factor analysis. It consists of
perpendicular projection of multi-parameter data on the plane represented by properly
selected eigenvectors V1 and V2, which are related to the highest eigenvalues of
covariance matrix of observational set. The selection of vectors V1 and V2 allows to
obtain an image on plane representing the greatest number of data changes whose
mutual distance is the biggest (Li et al., 2000).

Algorithm
A set of input data consists of parts described by n of features. It can be then treated as
a set of n-dimensional vectors. Let us mark the vector of input data as xk=(xk,1, xk,2, …
xk,n). The algorithm performing visualization by means of PCA consists of several
steps:
 scaling of initial data. Individual features represented by individual data
dimensions are scaled in the way ensuring their compliance with the same selected
range. In this paper, the individual coordinates (features) of data set vectors were
scaled to the range (0, 1),
578 T. Niedoba

Table 1. Data for size fractions of 14.00-12.50 mm for all three types of coal

Coal type 31
Ash Volatile parts
Density Mass Combustion heat Sulfur Analytical
contents contents
[Mg/m3] [g] [kJ] contents [%] moisture Wa
[%] Va
<1.3 308.6 29.51 6.41 0.72 34.32 3.23
1.3–1.4 292.5 24.53 19.61 0.7 29.22 3.36
1.4–1.5 36.1 12.34 16.55 0.76 28.92 3.87
1.5–1.6 10.7 21.42 26.10 1.55 31.08 3.40
1.6–1.7 25.6 18.70 35.78 2.28 26.71 2.40
1.7–1.8 139 16.41 37.20 1.23 29.24 2.19
1.8–1.9 12.7 12.89 48.20 1.13 24.05 2.23
>1.9 601.2 1.91 86.53 0.40 9.30 0.91
Coal type 34.2
Ash Volatile parts
Density Mass Combustion heat Sulfur Analytical
contents contents
[Mg/m3] [g] [kJ] contents [%] moisture Wa
[%] Va
<1.3 360.5 34.44 2 0.32 28.96 1.04
1.3–1.4 57 32.02 7.67 0.71 24.16 1.87
1.4–1.5 25.5 28.89 15.33 0.83 24.58 1.34
1.5–1.6 12.2 24.27 33.73 0.17 27.85 0.95
1.6–1.7 3.2 20.22 34.3 0.34 no data no data
1.7–1.8 15 17.38 36.15 0.34 27.93 0.37
1.8–1.9 3.6 18.48 27 0.05 31.75 1.01
>1.9 68.9 2.90 79.33 0.91 12.08 0.52
Coal type 35
Ash Volatile parts
Density Mass Combustion heat Sulfur Analytical
contents contents
[Mg/m3] [g] [kJ] contents [%] moisture Wa
[%] Va
<1.3 268.7 34.86 2.38 0.28 20.28 1.45
1.3–1.4 89.3 31.86 8.97 0.36 20.10 1.21
1.4–1.5 39.8 27.49 19.61 0.56 18.83 1.28
1.5–1.6 22.0 21.06 35.68 0.39 16.22 1.32
1.6–1.7 25.7 20.88 34.62 1.26 19.42 1.47
1.7–1.8 29.0 19.21 40.60 0.38 18.86 1.61
1.8–1.9 28.1 13.76 52.24 1.14 17.95 1.51
>1.9 589.5 2.94 80.57 0.20 10.84 1.37
 Multi-parameter data visualization by means of principal component analysis… 579

 calculation of covariance matrix. The general formula for covariance was used:

cov X , Y   E  X  Y   E  X   E Y  (1)

where E is the expected value. First, the expected values are calculated:
m

x
k 1
k ,i
Ei  (2)
m

and
m

x
k 1
k ,i x k , j
Ei , j  (3)
m

where, Ei – expected value of ith coordinate of input data, Ei,j – expected value of
product of ith and jth coordinate of input data, m – number of vectors of input data, xk,i
– ith coordinate of kth vector of input data. If the covariance matrix is marked as A, then
each element of the matrix aij is calculated as:

aij  Ei , j  Ei E j . (4)

In this way, the symmetrical covariance matrix of input data set is obtained,
 calculation of eigenvalues and eigenvectors of covariance matrix. For numerical
calculations the Jacobi method was selected. It draws upon the fact that orthogonal
transformation does not change own values and vectors of matrix. Consequently, it
is possible to perform a set of orthogonal transformations on matrix A to conduct it
to diagonal form D:

A  W  D W T . (5)

In the diagonal matrix the eigenvalues are located on the main diagonal while the
related eigenvectors are located in columns of matrix W. Matrices D and W which
fulfill Eq. 5 by means of the Jacobi method can be achieved in the following steps:
1. as matrix W the identity matrix of dimension nxn is accepted,
2. as matrix A covariance matrix of dimension nxn is accepted, calculated according
to equation (4),
3. the main element is selected apart from the main diagonal of matrix A, which is the
element of the highest module value not located on the main diagonal. The location
of this element in the matrix is searched, so coordinates p and q must be found
fulfilling the condition:
580 T. Niedoba

i, j  1,..., n and i  j is : a pq  aij , (6)

4. the values c and s are calculated using the following formulas:

a qq  a pp
r (7)
2a pq

sgn( r )
t (8)
r  r 2 1

where: aij is an element of the matrix from ith line and jth column, sgn(r) = 1 for r>=0
and sgn(r) = –1 for r < 0. Next, the calculations are performed according to formulas
(9) and (10):

1
c (9)
t 1 2

and
s  tc (10)

1. having applied calculated values c and s, matrix B is created ensuring identity

matrix of dimension nxn in which four elements are changed: bpp = c, bqq = c, bpq =
s and bqp = -s,
2. the new value of matrix A is assigned by application of previous value of matrix A,
matrix B created in the previous step and transposition matrix B:

A : B T  A  B , (11)

3. the new value of matrix W is assigned by application of previous value of matrix W

and matrix B created in step 5.
W : W  B , (12)

4. it is checked whether the matrix A, obtained as a result of transformations, is a

diagonal matrix according to previously accepted preciseness of calculations  , so:

max aij
i , j 1,..., n  i  j
 . (13)
max aii
i 1,..., n

For experiments described in section 3,   0.000001 was accepted. If inequality

(13) is not fulfilled, a return to step 3 is required and calculations are continued. In
 Multi-parameter data visualization by means of principal component analysis… 581

other case, the obtained matrix A is diagonal. The situation described by Eq. 5 is
achieved in this way. That means that the eigenvalues of output matrix are located on
the main diagonal of obtained matrix A and the eigenvectors related to these values
occur in columns of obtained matrix W,
 the determination of two coordinate axes. Two eigenvectors related to two biggest
module eigenvalues of covariance matrix are selected from vectors calculated in
stage C. Let us mark them as V1 = (v1,1, v1,2, … v1,n), V2 = (v2,1, v2,2, … v2,n). In this
way two coordinates axes are obtained on which all data will be projected,
 drawing the set of points on screen. For each point xk two coordinates are
calculated ( ~
xk ,1 , ~
xk , 2 ) preceded by their projection on axes V1 and V2, which
means:
n
~
x k ,1  v
i 1
1,i x k ,i (14)

n
~
xk , 2  v
i 1
2 ,i x k ,i . (15)

As a result, the image of each vector can be presented on computer screen. It is

achieved by drawing on the screen a symbol, in place of coordinates ( ~ xk ,1 , ~
xk , 2 ) ,
representing fraction to which the related data vector xk belongs. In this way the image
of multidimensional points representing various coal types is created on the computer
screen.

Results of experiment
As part of the investigation, for the purpose of visualizing seven-parameter data
describing various coal types, a computer program was created based on assumptions
presented in the previous chapter. The obtained results were presented in Figs 1-4. The
views show a transformation of seven-parameter data by means of PCA into two
dimensions. The visualization algorithm in PCA, despite reduction to only two
dimensions, has been organized to ensure the view representing the biggest number of
data changes whose mutual distance is the biggest. In this way it is possible to see
significant features of seven-parameter data on the computer screen.
582 T. Niedoba

Fig. 1. View of seven-parameter data representing three various types of coal.

Images of points representing coal, type 31 were marked with (■),
(+) – samples of coal type 34.2, (o) – samples of coal type 35.

Figure 1 presents obtained view of points representing seven-parameter data

vectors that describe three various coal types: 31, 34.2 and 35. In order to obtain it, the
system calculated the covariance matrix:

0.0834  0.0101  0.0773 0.0751 0.0162  0.0419  0.0151 

 0.0101 0.0241 0.0057  0.0056 -0.0049 0.0049 0.0062 
 
 0.0773 0.0057 0.0810 -0.0771 -0.0135 0.0418 0.0079 
 
cov  0.0751  0.0056  0.0771 0.0770 0.0097  0.0473  0.0148  .
0.0162  0.0049  0.0135 0.0097 0.0438 0.0074 0.0150 
 
 0.0419 0.0049 0.0418  0.0473 0.0074 0.0545 0.0245 
 0.0151 0.0062 0.0079  0.0148 0.0150 0.0245 0.0396 
 

Furthermore, eigenvectors were calculated:

V1 = (0.5401, -0.0589, -0.5340, 0.5307, 0.0763, -0.3425, -0.1225),
V2 = (0.1289, 0.0025, -0.1631, 0.0202, 0.6594, 0.4273, 0.5822).
 Multi-parameter data visualization by means of principal component analysis… 583

In the above Figure, it is clearly visible that images of data points representing coal
samples of a certain type gather in separated subareas and create clusters. It can be
noticed that almost on the whole area of the Figure these clusters can be separated. In
some parts of the space, however, the images of points representing various coal types
overlap. Therefore, according to Figure 1, it is impossible to state that the analyzed
data allow to classify coal types properly.

Fig. 2. View of seven-parameter data representing two various types of coal. Images of points
representing coal type 31 were marked with (■), (o) – samples of coal type 35.

With a view to achieving clearer results, it was decided to present this data by
means of the PCA method in some other way. A decision was made to analyze the
data representing various types of coal in pairs. Figure 2 shows the view obtained for
data representing coal types 31 and 35. In order to obtain this view, the system
calculated the covariance matrix:
584 T. Niedoba

0.1071 - 0.0115 - 0.0827 0.0842 0.0214 - 0.0466 - 0.0261 

- 0.0115 0.0305 0.0053 - 0.0056 - 0.0077 0.0068 0.0056 

- 0.0827 0.0053 0.0731 - 0.0719 - 0.0144 0.0351 0.0149 
 
cov  0.0842 - 0.0056 - 0.0719 0.0746 0.0095 - 0.0445 - 0.0251  .
0.0214 - 0.0077 - 0.0144 0.0095 0.0508 0.0159 0.0152 
 
- 0.0466 0.0068 0.0351 - 0.0445 0.0159 0.0564 0.0453 
- 0.0261 0.0056 0.0149 - 0.0251 0.0152 0.0453 0.0498 

Furthermore, eigenvectors were calculated:
V1 = (0.5936, -0.0637, -0.4798, 0.5031, 0.0701, -0.3325, -0.2116),
V2 = (0.1948, -0.0316, -0.1952, 0.0507, 0.6128, 0.4902, 0.5519).
In Figure 2 it is clearly visible that images of points representing samples of coal
type 31 gather in clusters, which can be easily separated from clusters of points
representing coal type 35. In Figure 3 the view obtained for data representing coal
types 34.2 and 35 is presented. Also, in this case it is clearly visible that images of
points representing samples of coal type 34.2 gather in clusters, which can be easily
separated from clusters of points representing samples of coal type 35. The covariance
matrix obtained from the creation of this Figure is as follows:

0.0840 - 0.0101 - 0.0851 0.0811 0.0111 - 0.0440 - 0.0021 

- 0.0101 0.0222 0.0059 - 0.0046 - 0.0083 - 0.0007 0.0007 
 
- 0.0851 0.0059 0.0922 - 0.0886 - 0.0059 0.0523 0.0028 
 
cov  0.0811 - 0.0046 - 0.0886 0.0868 0.0055 - 0.0542 - 0.0005  .
0.0111 - 0.0083 - 0.0059 0.0055 0.0278 - 0.0010 0.0015 
 
- 0.0440 - 0.0007 0.0523 - 0.0542 - 0.001 0 0.0650 - 0.0143 
- 0.0021 0.0007 0.0028 - 0.0005 0.0015 - 0.0143 0.0374 
 
Its eigenvectors are as following:
V1 = (-0.5174, 0.0414, 0.5554, -0.5404, -0.0480, 0.3575, -0.0090),
V2 = (-0.2219, 0.1868, 0.1310, -0.0412, -0.1434, -0.6099, 0.7102).
Furthermore, Fig. 4 shows the view obtained for data representing coal types 31
and 34.2. It can be easily noticed that images of points representing samples of coal
type 31 gather in clusters which can be easily separated from clusters of points
representing samples of coal type 34.2. In this case the covariance matrix is as
follows:
 Multi-parameter data visualization by means of principal component analysis… 585

0.0840 - 0.0121 - 0.0763 0.0745 0.0208 - 0.0494 - 0.0246 

- 0.0121 0.0256 0.0080 - 0.0081 - 0.0037 0.0080 0.0103 

- 0.0763 0.0080 0.0805 - 0.0770 - 0.0178 0.0518 0.0149 
 
cov  0.0745 - 0.0081 - 0.0770 0.0778 0.0149 - 0.0551 - 0.0227  .
0.0208 - 0.0037 - 0.0178 0.0149 0.0536 - 0.0020 0.0161 
 
- 0.0494 0.0080 0.0518 - 0.0551 - 0.0020 0.0513 0.0254 
- 0.0246 0.0103 0.0149 - 0.0227 0.0161 0.0254 0.0543 


The eigenvectors are:

V1 = (-0.5243, 0.0776, 0.5169, -0.5167, -0.1134, 0.3743, 0.1788),
V2 = (0.0762, 0.0760, -0.1317, 0.0184, 0.7040, 0.1959, 0.6608).

Fig. 3. View of seven-parameter data representing three various types of coal. Images of points
representing coal type 34.2 were marked with (+), (o) – samples of coal type 35.

If we can state that the distinction of coal type 31 samples from coal type 35
samples is possible (Fig. 2) and we can state that the distinction of coal type 34.2
586 T. Niedoba

samples from coal type 35 samples is possible (Fig. 3), as well the distinction of coal
type 31 samples from coal type 34.2 samples is possible (Fig. 4), then the possibility
of distinction of samples of each of the three coal types is confirmed. Thanks to the
visualization of multi-parameter data by means of PCA, it is possible to state that
information covered in seven-parameter data describing samples of three types of coal
is sufficient for its proper classification.
It is worth paying attention to the fact that the algorithm of visualization by means
of PCA does not use the information of affiliation of points representing data to
certain fractions. In this situation grouping of points representing certain fraction
depends only on some properties of data noticed by the algorithm.

Fig. 4. View of seven-parameter data representing two various types of coal. Images of points
representing coal type 31 were marked with (■), (+) – samples of coal type 34.2

Conclusions
The conducted experiments based on visualization of seven-parameter data by means
of PCA allowed to arrive at the following conclusions.
 Multi-parameter data visualization by means of principal component analysis… 587

1. Multi-parameter visualization by means of the PCA allows to state that information

covered in analyzed seven-dimensional data is sufficient for a proper classification
of coal types 31, 34.2 and 35.
2. The visualization of data concerning three types of coal within one figure allowed
to state that images of data points representing coal samples of certain type gather
in clusters which can be separated almost on the whole area of the figure. However,
in some areas of the space, the images of points representing various coal types
overlap. Therefore, based on such view it was impossible to state whether the
analyzed data allows for a proper classification of coal types.
3. Only presentation of data representing three various types of coal in pairs allowed
to obtain clear results. They allowed to conclude that images of points representing
samples of coal of certain type gather in clusters which can be separated. It means
that data contains a sufficient amount of information to classify coal types
properly.
4. The advantage of the PCA method is the fact that during visualization it is not
necessary to select any parameters in contrast to many other methods of multi-
parameter data visualization.

Acknowledgements
This paper is the result of a scientific project no. N N524 339040, agreement no. 3390/B/T02/2011/40.

References
AHMED H.A.M., DRZYMALA J., Two-dimensional fractal linearization of distribution curves,
Physicochemical Problems of Mineral Processing, vol. 39, pp. 129-139, 2005.
ALDRICH C., Visualization of transformed multivariate data sets with autoassociative neural networks.
Pattern Recognition Letters, Volume: 19, Issue: 8, June, 1998, pp. 749-764.
ASIMOV D., The Grand Tour: A Tool for Viewing Multidimensional Data, SIAM Journal of Scientific
and Statistical Computing, pp. 128-143, vol. 6, No, 1985.
ASSA J., COHEN-OR D., MILO T., RMAP: a system for visualizing data in multidimensional relevance
space, Visual Computer, vol.15, no.5, pp.217-34. Publisher: Springer-Verlag, Germany, 1999.
BROZEK M., SUROWIAK A., Argument of Separation at Upgrading in the Jig, Archives of Mining
Sciences, vol. 55, pp. 21-40, 2010.
BROZEK M., SUROWIAK A., Effect of Particle Shape on Jig Separation Efficiency, Physicochemical
Problems of Mineral Processing, vol. 41, pp. 397-413, 2007.
BROZEK M., SUROWIAK A., The Dependence of Distribution of Settling Velocity of Spherical
Particles on the Distribution of Particle Sizes and Densities, Physicochemical Problems of Mineral
Processing, vol. 39, pp. 199-210, 2005.
CHATTERJEE A., DAS P.P., BHATTACHARYA S., Visualization in linear programming using
parallel coordinates, Pattern Recognition, vol. 26(11), pp.1725-1736, 1993.
CHOU S.-Y., LIN S.-W., YEH C.-S., Cluster identification with parallel coordinates, Pattern
Recognition Letters, vol. 20, pp.565-572, 1999.
CLEVELAND W.S., MCGILL R., The many faces of a scatterplot, Journal of the American Statistical
Association, vol.79, pp.807-822. 1984.
588 T. Niedoba

COOK D., BUJA A., CABRERA J., HURLEY C., Grand Tour and Projection Pursuit, Journal of
Computational and Graphical Statistics, vol. 4, no. 3, pp. 155-172, 1995.
DRZYMALA J.: Basics of minerallurgy, Oficyna Wydawnicza Politechniki Wrocławskiej, 2009. [in
Polish]
GAWENDA T., SARAMAK D., TUMIDAJSKI T., Regression models of rock materials crushing in jaw
crushers, Scientific Issues of Civil Engineering and Environmental Engineering Faculty of Koszalin
University of Science and Technology, series: Environmental Engineering, no 22, pp. 659-670, 2005.
[in Polish]
GENNINGS C., DAWSON K.S., CARTER W.H., JR. MYERS R.H., Interpreting plots of a
multidimensional dose-response surface in a parallel coordinate system. Biometrics, vol. 46, pp.
719-735, 1990.
INSELBERG A., The plane with parallel coordinates. Visual Computer, vol. 1, pp. 69-91, 1985.
JAIN A.K., MAO J., Artificial neural network for non-linear projection of multivariate data. In: Proc.
IEEE Internat. Joint Conf. On Neural Networks, Baltimore, MD, vol. 3, pp. 335-340, 1992.
JAMROZ D., NIEDOBA T., Application of Observational Tunnels Method to Select Set of Features
Sufficient to Identify a Type of Coal, Physicochemical Problems of Mineral Processing, vol 50(1), pp.
185-202, 2014.
JAMROZ D., Visualization of objects in multidimensional spaces, Doctoral Thesis, AGH, Kraków, 2001.
[in Polish]
JAMROZ D., Application of Multidimensional Data Visualization in Creation of Pattern Recognition
Systems, In: Gruca A., Czachórski T., Kozielski S. (eds.), Man-Machine, Interactions 3, AISC,
Heidelberg, Springer-Verlag, vol. 242, pp. 443-450, 2014.
JAMROZ D.: Multidimensional labyrinth - multidimensional virtual reality. In: Cyran K., Kozielski S.,
Peters J., Stanczyk U., Wakulicz-Deja A. (eds.), Man-Machine, Interactions, AISC, Heidelberg,
Springer-Verlag, pp. 445–450, 2009.
KELLY E.G., SPOTTISWOOD D.J., The Theory of Electrostatic Separations, a Review, Part I.
Fundamentals, Minerals Engineering, vol. 2/1, 33-46, 1989.
KIM S.-S., KWON S., COOK D., Interactive visualization of hierarchical clusters using MDS and MST.
Metrika, vol. 51, Springer-Verlag, pp. 39-51, 2000.
KRAAIJVELD M., MAO J., JAIN A.K., A nonlinear projection method based on Kohonen’s topology
preserving maps. IEEE Trans. Neural Networks, vol. 6(3), pp. 548-559, 1995.
LI W., YUE H. H., VALLE-CERVANTES S., QIN S. J., Recursive PCA for adaptive process
monitoring. Journal of Process Control, vol. 10, iss.: 5, pp. 471-486, 2000.
LYMAN G. J., Application of Line-Length Related Interpolation Methods to Problems in Coal
Preparation – III: Two dimensional Washability Data Interpolation, Coal Preparation, vol. 13, pp.
179-195, 1993.
NIEDOBA T., Multidimensional characteristics of random variables in description of grained materials
and their separation processes, Wydawnictwo Instytutu Gospodarki Surowcami Mineralnymi i
Energią PAN, Kraków, 2013a [in Polish].
NIEDOBA T., Multidimensional distributions of grained materials characteristics by means of non-
parametric approximation of marginal statistical density function, AGH Journal of Mining and
Geoengineering, iss. 4, pp. 235-244, 2009 [in Polish].
NIEDOBA T., Statistical analysis of the relationship between particle size and particle density of raw
coal, Physicochemical Problems of Mineral Processing, vol. 49, iss. 1, pp. 175-188, 2013b.
 Multi-parameter data visualization by means of principal component analysis… 589

NIEDOBA T., JAMROZ D.: Visualization of multidimensional data in purpose of qualitative

classification of various types of coal, Archives of Mining Sciences, vol. 58, iss. 4, pp. 1317-1333,
2013.
NIEDOBA T., SUROWIAK A., Type of coal and multidimensional description of its composition with
density and ash contents taken into consideration, in Proceedings of the XXVI International Mineral
Processing Congress, vol. 1, pp. 3844-3854, 2012.
NIEDOBA T., Three-dimensional distribution of grained materials characteristics, in Proceedings of the
XIV Balkan Mineral Processing Congress, Tuzla, Bosnia and Herzegovina, vol. 1, pp. 57-59, 2011.
OLEJNIK T., SUROWIAK A., GAWENDA T., NIEDOBA T., TUMIDAJSKI T., Multidimensional coal
characteristics as basis for evaluation and adjustment of its beneficiation technology, AGH Journal of
Mining and Geoengineering, vol. 34, iss. 4/1, pp. 207-216, 2010. [in Polish]
SARAMAK D., Mathematical models of particle size distribution in simulation analysis of High-pressure
grinding rolls operation, Physicochemical Problems of Mineral Processing, vol. 49(1), pp. 495–512,
2013.
SARAMAK D., Technological Issues Of High-Pressure Grinding Rolls Operation In Ore Comminution
Processes, Archives of Mining Sciences, vol. 56, no 3, pp. 517-526, 2011.
SNOPKOWSKI R., NAPIERAJ A., Method Of The Production Cycle Duration Time Modeling Within
Hard Coal Longwall Faces, Archives of Mining Sciences, vol. 57, no. 1, pp. 121-138, 2012.
SOBOL M.G., KLEIN G., New graphics as computerized displays for human information processing.
IEEE Trans. Systems Man Cybernet., vol. 19 (4), pp. 893-898, 1989.
TUMIDAJSKI T., SARAMAK D., Methods and models of mathematical statistics in mineral processing,
Wydawnictwo AGH, Kraków, 2009. [in Polish]
TUMIDAJSKI T., Stochastic analysis of grained materials properties and their separation processes,
Wydawnictwo AGH, Kraków, 1997 [in Polish].
590 T. Niedoba