Journal of Physics: Conference Series

PAPER • OPEN ACCESS You may also like

- English Translation Effect Evaluation
Analysis of machine learning methods to improve System Based on Big Data
Wang Xudong
efficiency of big data processing in Industry 4.0 - Structural design of anthropomorphic robot
vision system
Junyan Zhang
To cite this article: A A Prudius et al 2019 J. Phys.: Conf. Ser. 1333 032065
- Analysis on Research Status of Road
Performance of Machine-made Sand
Concrete
Luo Ting-yi, Tang Ya-sen, Xu Peng et al.
View the article online for updates and enhancements.

This content was downloaded from IP address 37.220.117.231 on 08/02/2023 at 13:04

ITBI 2019 IOP Publishing
Journal of Physics: Conference Series 1333 (2019) 032065 doi:10.1088/1742-6596/1333/3/032065

Analysis of machine learning methods to improve efficiency of

big data processing in Industry 4.0

A A Prudius1, A A Karpunin1, A I Vlasov1

1
Bauman Moscow State Technical University, 5/1, 2-nd Baumanskaya Str., Moscow,
105005, Russian Federation

Abstract. The article considers the basic methods of machine learning applied by individual
entrepreneurs within the framework of transition to digital production to improve the efficiency
of data processing, classification of existing and would-be customers and their subsequent work
with them. The main attention is paid to the problem of increasing the effectiveness of methods
of machine learning applied for solving the current questions. Areas of application of technology
are shown. The peculiarities of machine learning are briefly analyzed. The main features and
prospects of the development of Machine Learning services are shown on the basis of the concept
of a step-by-step combination of the methods under consideration. One of the main algorithms
for working with data is analyzed; its main features, scope and procedure are described.
Recommendations are given for the further use of machine learning algorithms. The role of
machine learning in the development of modern science and industry is analyzed, the main
tendencies of the industry development are determined, and the practical application of big data
is shown. As part of the transition to Industry 4.0, the main areas of application of machine
learning, big data, Artificial Intelligence and their relations with the corresponding fields of
science and production are described. The article also offers a review of the application of
Artificial Intelligence and machine learning in particular in the context of the transition to
digitalization and the issues of individual entrepreneurship.

1. Introduction
The issue of analyzing the processing of large amounts of data has to be addressed starting from the
earliest stages in the development of any applied task. Transitioning the development of modern industry
to Industry 4.0 [1], which primarily implies the implementation of cybernetics and AI (Artificial
Intelligence) and its components in different processes, which includes raising radically new questions
in development. So today implementing machine learning and using big data are an integral part of
efficient work in every organization.
Machine Learning (ML) is one of the most relevant directions in the development of modern
technology, which represents methods of artificial intelligence aimed at training systems in the course
of practical resolution of a number of applied tasks. When considering machine learning as a task, the
task can be formulated as a generalization of classical approximation tasks, since in real-life of the
applied projects the data can be non-numeric, incomplete and heterogeneous. Meanwhile, many machine
learning methods are closely related to retrieving information and data mining [2].
The relevance of this research stems from the fact that the extensive growth of information volumes
in recent times results in the emergence of Big Data as a standalone class of information processed,

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
ITBI 2019 IOP Publishing
Journal of Physics: Conference Series 1333 (2019) 032065 doi:10.1088/1742-6596/1333/3/032065

which requires using networked computing capacity to process, as it is beyond the human mind to handle
these amounts of data.
Machine learning currently has a broad range of applications [3]. The applications of machine
learning are expected to continue growing due to ubiquitous digitization and accumulation of huge
amounts of data in science, industry, transportation and business sectors of the society [4].
The work offers a methodology to analyze customer base in financial and insurance organizations,
in order to classify existing and prospective clients using several machine learning algorithms. Its
scientific novelty comes from using the sequence of information processing and applying the most
efficient combinations of different methods.
The results of the study are practically relevant, as they increase the precision of the data obtained
during the analysis at each level, which can further be applied in building corporate behavior strategies.

2. Literature review
This work reviews the issue of analyzing the efficiency of data analysis and applying radically new
methods of dealing with data. Statistical analysis methods are used worldwide, the only thing that
changes with time is the technology, which is currently based on machine learning. By using advanced
analytics, financial organizations obtain practical applications of universal statistical methods and as a
result gain valuable insights on their customer base, peculiarities of customer behavior and causal
relationships in their interaction with the business.
Machine learning generally is one of the most common methods of research used by artificial
intelligence. In the process of ubiquitous implementation of cybernetic systems in various areas of
human life and developing Industry 4.0, AI plays one of the most important roles in the development of
modern science and industry, thanks to the broad range of applications [5].
Some of the key practical applications of artificial intelligence include such trendy technologies as
BigData [6], Internet of Things (IoT) [7], Virtual Reality/Augmented Reality (VR, AR) [8], and
distributed ledgers (Blockchain) [9-12]. These and many others help to form a new stage in the
development of public conscience and scientific development, directed to a broad range of theoretical
and practical tasks addressed on a daily basis.

3. Methods
The key methods of machine learning [10] examined in this work for addressing the tasks at hand are as
follows:
Clustering, the goal of which is to break down a multitude of various objectives into groups (called
“clusters”) that contain “similar” objects with the maximum possible differences between the clusters.
Clustering is the most general method of machine learning of those considered in this article; its main
difference from classification is that the list of groups resulting from the breakdown is not fixed and is
defined in the process of work (unsupervised learning) [11];
Classification, which is required to break down a finite number of objects into predefined classes by
one or more parameters (supervised training);
Regression, which involves studying the impact of one or more independent variables on the
dependent variable. Regression is the most accurate data analysis method.
One of the most common algorithms in machine learning, as used in addressing applied tasks, is the
logistic regression of input parameters, which is characterized by lack of predicted value of numeric
variables from the sampling of input data [13]. Instead of returning a certain numeric value, the function
determines the probability of the value being analyzed matching a certain class of system parameters.
For clarity, let us suppose only two classes are being analyzed, and the probability to be determined, P+,
is the probability of a certain value belonging to the class “+”. Respectively, P– = 1 – P+for all other
cases. Thus, the output of logistic regression is always in the interval of [0, 1].
The idea of logistic regression is that a multitude of input parameters can be separated by a plane (in
case of two dimensions – with a straight line with no inflection points) into two areas corresponding to
the classes of output data. The resulting (see Fig. 1) plane is called ‘linear discriminant’, as from the

2
ITBI 2019 IOP Publishing
Journal of Physics: Conference Series 1333 (2019) 032065 doi:10.1088/1742-6596/1333/3/032065

viewpoint of its function it is linear and allows the input parameters to be discriminated into different
classes [14].

Figure 1. Visual representation of discrimination by logistic regression

In the simplest case, when considering a two-dimensional space, for a point with coordinates (a, b)
the logistic regression algorithm will include the following steps:
1. Calculate the value β0+β1a+β2b
of the boundary function (or odds ratio function) identified as t;
2. Calculate the odds ratio: OR+ = et (as t is a logarithm);
OR
3. Once the value OR+ is obtained, calculate P+ using a simple dependency: P  .
1  OR
et
After getting the value t in p.1 and combining the remaining steps, we get P  . The formula to
1  et
the right of the equation is the logistic function, which lends its name to the method of analysis.
By combining this algorithm with such ML methods as gradient boosting and random forest, we can
improve the accuracy of input value classification, increasing precision of data analysis.
The implementation of algorithms of machine learning is possible in virtually any programming
language. The most popular are the following ones: Python; R; Scala; Java.
Each of the listed programming languages implies the availability of its own tools for working with
machine learning algorithms. In the framework of this paper, it is proposed to consider some libraries
of the Python language as the most common ones.
One of the main libraries is NumPy; it is a Python library that adds support for large multidimensional
arrays and matrices together with a large library of high-level mathematical functions for operations
with these arrays.
Mathematical algorithms implemented in Python often work much slower than the same algorithms
implemented in compiled languages (for example, Fortran, C, Java). The NumPy library provides
implementations of computational algorithms (in the form of functions and operators), optimized for
working with multidimensional arrays. As a result, any algorithm that can be expressed as a sequence
of operations on arrays of data and implemented using NumPy is equivalent in speed to the code
processed in the MATLAB environment.
To visualize the results of work, libraries such as Matplotlib, Bokeh, Plotly, Scikit are used. For
example, Matplotlib is a Python library for building qualitative two-dimensional graphs. Matplotlib is a
flexible, easily configurable package that along with the main language libraries provides capabilities
similar to the MATLAB environment. At present, the package works with several graphics libraries,
including wxWindows and PyGTK. Simple three-dimensional graphics can be built using the toolkit

3
ITBI 2019 IOP Publishing
Journal of Physics: Conference Series 1333 (2019) 032065 doi:10.1088/1742-6596/1333/3/032065

mplot3d. There are other sets of tools: for cartography, for working with Excel, utilities for GTK and
others.
There is also a class of libraries specially designed to work with machine learning, including Theano,
TensorFlow, Keras, etc. Theano is an extension of the Python language that allows you to efficiently
compute mathematical expressions containing multidimensional arrays. The library provides a basic set
of tools for the configuration of neural networks and their learning. Theano has received the greatest
recognition in the tasks of machine learning when solving optimization problems. The library allows
you to use the GPU features without changing the program code, which makes it irreplaceable when
performing resource-intensive tasks.

4. Results and Discussion

In practice, solving applied tasks by using machine learning methods as part of this work is reduced to
the joint usage of the algorithms described above.
The first issue to address is data clustering. To do this, all characteristic parameters are put in an N-
dimensional space, where N is the number of dimensions for, and the distance between anchor points
(which represent customer characteristics) is calculated. This way the entire volume of data is segmented
by the criteria needed in each specific case, for further targeted work with each group. As part of the
first stage of data processing, we also segment input data based on a custom quantity of parameters
being analyzed and the most detailed segmentation by the maximum number of factors, depending on
the goals set. The peculiarity of this stage lies in obtaining the results of statistical analysis without
indicating priority parameters for further data work.
When the resulting data are classified based on the correlation between data points and a zero
variable, the factors of the information analyzed are weighed (e.g., paying capacity has the maximum
correlation and is the target variable when analyzing paying capacity), based on which each segment is
fit within one class or another. Multi-collinearity (see Fig. 2) also plays a role – when using multiple
factors, their individual contribution to classification can change substantially, in presence of a strong
correlation between the factors – even though an individual factor can have substantial weight, it can be
reduced in combination with additional factors with high correlation.

Figure 2. Results of correlation analysis of the variables

Additionally, more efficient algorithms can be used to optimize performance during the data
classification phase – e.g., gradient boosting, which is based on multiple decision trees and can learn
from its own mistakes – when building each subsequent decision tree (see Fig. 3), it considers and
corrects errors made in earlier iterations.

4
ITBI 2019 IOP Publishing
Journal of Physics: Conference Series 1333 (2019) 032065 doi:10.1088/1742-6596/1333/3/032065

Figure 3. Building a decision tree

Finally, regression analysis methods are applied, which, unlike classification, help identify specific
values of different factors, such as choosing an interest rate that is the most profitable for the company
that still meets the customer requirements, considering all the factors and aspects of the given situation.
Even without having specific values but being aware of all the parameters, one can give a highly accurate
estimation of the figure that can be adapted for each specific case.
Thus, by using all methods of machine learning in sequence, an organization gets a highly detailed
analysis of its customer base, which enables highly targeted work with different customer groups and
each individual customer. This method is highly optimized for the business, by determining specific
values on a case-by-case basis, following the overall targets formulated by the company.

5. Conclusion
The work studies key methods of machine learning, their peculiarities and application areas. The authors
review algorithms for analyzing big data using ML, identify their optimum combinations for
personalizing the approach to individual customers, improve the efficiency of a financial organization
(e.g., reducing costs, improving conversion and margins) and optimizing and automating business
processes within companies. It can be concluded that analyzing ordinary individual events in the context
of big data improves the performance of an individual company operation and the industry in general,
capturing the benefits that could not be accessed previously.

6. Acknowledgement
Some results of the work were obtained with the support of project No. 14.579.21.0142 UID
RFMEFI57917X0142 within the framework of the federal target program “Research and development
in priority areas of development of the scientific and technological complex of Russia for 2014-2020”.

References
[1] Akberdina V V, Tretyakova O V and Vlasov A I 2017 A methodological approach to
forecasting spatial distribution of workplaces in an industrial metropolis Problems and
Perspectives in Management 4(87) 50–61

5
ITBI 2019 IOP Publishing
Journal of Physics: Conference Series 1333 (2019) 032065 doi:10.1088/1742-6596/1333/3/032065

[2] Hastie T, Tibshirani R and Friedman J. 2009 The Elements of Statistical Learning: Data
Mining, Inference, and Prediction (Springer-Verlag) p 746
[31] Vlasov A I and Yuldashev M N 2018 Analysis of the methods and tools for processing
information from a cluster of sensor Sensors and Systems 1(221) 24–30
[4] Vlasov A I, Yudin A V, Shakhnov V A, Usov K A and Salmina M A 2017 Design methods of
teaching the development of internet of things components with considering predictive
maintenance on the basis of mechatronic devices International Journal of Applied Engineering
Research 12(20) 9390–9396
[5] Witten I H and Frank E 2016 Data Mining: Practical machine learning tools and techniques
Morgan Kaufmann 7–9
[6] Vlasov A I, Novikov P V, and Rivkin A M 2015 Peculiarities of airtraffic planning using
weather maps built with BIG DATA technologies Bulletin of the Bauman Moscow State
Technical University. Series: Machine Building 6(105) 46–62
[7] Vasconcelos D R 2017 Mathematical model for a collaborative indoor position system (IPS)
and movement detection of devices within IoT environment Proceedings of the Symposium on
Applied Computing 602–608.
[8] Muravyev K A, Averyanikhin A E and Kotelnitsky A V 2016 A methodology of calculating the
optimal number of nodes in a virtualization cluster for a private virtual desktop cloud for
maximum efficiency International Research Journal 5-3(47) 6–13
[9] Vlasov A I, Karpunin А А and Novikov I P 2017 Systems analysis of blockchain data exchange
and storage technology Modern technology. Systems Analysis. Modeling 3(55) 75–83
[10] Berduygina O N, Vlasov A I and Kuzmin E A 2017 Investment capacity of the economy during
the implementation of projects of public-private partnership Investment Management and
Financial Innovations 14(3) 189–198
[11] Yuldashev M N, Vlasov A I and Novikov A N 2018 Energy-efficient algorithm for
classification of states of wireless sensor network using machine learning methods Journal of
Physics: Conference Series 1015 032153
[12] Karpunin A A and Kozlov A A 2017 Analysis of the methods for implementing decentralized
applications in design and engineering information technology Information technology in
design and manufacturing 4(168) 39–44
[13] Vlasov A I and Yuldashev M N 2017 Gaussian processes in regression analysis of the states of
wireless sensor network, including electromagnetic interference Electromagnetic compatibility
technologies 3(62) 35–43
[14] Andreev K A, Vlasov A I and Shakhnov V A 2016 Silicon pressure transmitters with overload
protection Automation and Remote Control 77(7) 1281–1285