Abstract:-This article addresses the challenges in the application of artificial intelligence methods such as machine
learning, computational intelligence and/or soft computing methods in social sciences. The literature review is
performed in order to give a review of different approaches and methods that have been applied so far. The most
used method in social sciences and management is the SWOT method, for the identification of strengths,
weaknesses, opportunities, and threats when making strategic decisions. Two fundamental characteristics of
previous approaches are the development of numerical models of utility functions and the possibility of upgrading
these models by formalizing the intuition of strategic decision-makers. There are several shortcomings of the
existing approaches. The application of computational intelligence and machine learning methods in social
sciences is identified as one of the most challenging and promising areas, which could overcome identified
shortcomings. The principles of one popular machine learning method, the decision tree, are explained and a
demonstration is performed on the case study of churn prediction. Benchmarking data set from the publicly
available repository is used to demonstrate the suggested approach Evaluation results measured through model
accuracy and reliability gave promising results for further analysis. A developed predictive model could serve as a
standalone tool or as support for decision-making in social sciences.
Key-words:-Computational intelligence, data mining, data science, machine learning, social sciences, business.
Received: April 14, 2022. Revised: January 9, 2023. Accepted: February 4, 2023. Published: March 7, 2023.
1 Introduction
This article seeks to contribute to solving one of the
most challenging problems of artificial intelligence:
application for planning in social sciences. Artificial
Intelligence becomes the key technology of the XXI
century, which will become the most significant
economic branch in the next decade, with a great
impact on all areas of human activity. Today, artificial
intelligence demonstrates its superiority in solving
well-structured lower intellectual-level issues such as
machine learning (e.g. business analytics), visual
recognition, speaking, translating, and converting text
into speech. Solving complex problems such as
planning in social sciences still needs to prove the
ability of artificial intelligence. Modern
business-based analytics, based on machine learning,
is successful in supporting decision-makers and
automating processes in the business of lower
intellectual levels. Despite the supremacy of artificial
intelligence concerning human beings in many areas,
strategic planning is still unmanageable for people of
their intuition and creativity, and the ability to see
long-lasting changes that are not yet predictable to
existing business analytics.Today's organizations,
from the smallest (SME) to the global ones, base their
actions on highly structured strategic planning
methods. One of these techniques, SWOT analysis, is
the most commonly applied management method. On
the other hand, some approaches completely deny the
applicability of strategic management, as they believe
business systems are operating in a very dynamic
environment where rigid planning can`t be long-term
effects. Some newly-created technological giants base
their business on employee creativity, and less on
long-term planning. Most organizations still operate
based on planning. Increased dynamism and speed of
change in modern business (shortening product
cycles) increasingly diminish the effectiveness of
strategic planning, while at the same time increasing
the need for it. Planning in social sciences (as a
general term) is one of the affirmed areas of artificial
intelligence. Decision makers are increasingly relying
on analytics, but only partially in creating a strategy
(e.g. 30%). Prediction based on historical data,
achieved through machine learning techniques, is the
basis for effective strategic planning. The existing
Application of Machine Learning Methods for Data Analytics in
Social Sciences
DIJANA ORESKI
Faculty of Organization and Informatics,
University of Zagreb,
Varazdin, Pavlinska 2,
CROATIA
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2023.22.8
Dijana Oreski
E-ISSN: 2224-2678
69
Volume 22, 2023
research was primarily concerned with the upgrading
of qualitative techniques of strategic planning by
quantitative methods (e.g. machine learning
techniques, expert systems, and intelligent agents). A
special challenge is to provide citizens with a rational
decision-making that has significant long-lasting
impacts on their lives. This research aims to prove the
applicability of artificial intelligence and machine
learning methods in social sciences. This paper is
structured as follows. Section 2 provides a review of
existing approaches. Section 3 explains used
methodology and data. Section 4 gives an overview of
the research results. Section 5 concludes the paper.
2 Related Literature
Strengths, Weaknesses, Opportunities, and Threats
(SWOT) analysis is a widely used technique and one
of the most common tools in management. SWOT is a
brief list of statements or factors with descriptions of
the present and future trends of both the internal and
external environment. However, SWOT analysis has
no means of determining the importance of each
SWOT factor [1]. Thus, the utilization of SWOT
alone in decision-making process is insufficient.
Kurttila et. al. [2] recognized this limitation of SWOT
analysis and its impreciseness of a quantitative
examination. They created a hybrid SWOT-AHP
method where SWOT analysis usability was
improved. The limitation of the qualitative nature of
SWOT analysis is then overcome with the quantitative
SWOT-AHP method, but they still both stayed
subjective, developed by the human decision-makers.
SWOT-AHP has been used for strategic planning [3]
in various domains, such as tourism [4] and
manufacturing [5] In 1999. Houben at al. [6] described
an interesting application of a knowledge-based
system (KBS) to SWOT-analysis strategic planning in
small and medium-sized enterprises. They are focused
on the identification of internal strength and weakness
factors recognized by this KBS from the financial
situation of an organization. There are only a few
papers so far that utilize the mainstream of a huge
database growth and wide application of business
intelligence and data mining to the definition of
organizational strategies with the common and
acceptable frame of SWOT analysis. Knowledge
Discovery in Databases (KDD) and Data Mining
(DM) techniques can model most complex systems
accurately outperforming previously established
linear methods. KDD and DM can develop models of
complex systems represented by neural networks and
decision trees. Furthermore, Milano et.al. [7] tried to
cover “public policy issues in a wide variety of fields:
economy, education, environment, health, social
welfare, and national and foreign affairs. They are
extremely complex, characterized by uncertainty, and
involve conflicts among different interests.” Authors
[7] also see the advantages of artificial intelligence as
a solution for such complex problems. Athey [8]
recognized big data potential in policy problems.
Based on the aforementioned, this paper seeks to use
the advantages of artificial intelligence and machine
learning in order to solve strategic decision-making
issues in social sciences.
3 Data and Methods
In the first two sections of the paper, we have
described the recent developments and applications of
strategic decision-making methods and artificial
intelligence. The literature review demonstrated that
present methods are insufficient for application in
social systems that are nonlinear, complex, and based
on complex dynamic laws, and variables in such
systems are often not possible to measure exactly.
This was the motivation for a new approach based on
the application of artificial intelligence methods. Our
design is based on the following models:
(i) Application of data mining and standard
methods for conducting CRISP-DM.
(ii) Simulations driven by goals and data.
(iii) Evaluation and interpretation of predictive
models.
The steps of the research based on the CRISP-DM
methodology are explained in table 1.
Table 1. Research description through phases
Steps
Assessment of the environment
Definition of business goals
Assessment of the situation
Determining performance criteria
Initial data collection
Data description
Basic statistical analysis
Data quality assessment
Model structure development
Data set description
Data selection
Data cleaning
Deriving attributes
Data integration
Choice of modeling technique
Definition of model parameters
Model description
Evaluation of data mining results in
relation to business success criteria
Model interpretation
Application activity plan
Implementation and performance
control
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2023.22.8
Dijana Oreski
E-ISSN: 2224-2678
70
Volume 22, 2023
The advantage of CRISP-DM is that combines the
development of models by applying data mining
techniques, and supplements the model with the
knowledge and intuition of past data on the given
topic and domain in social sciences.
4 Research Results
CRISP-DM standard is applied to data from one
domain of social sciences, business. To demonstrate
the application of machine learning in social science
datasets about predicting whether a customer will
change telecommunications provider, something is
known as "churning", is used. The source of the
dataset is the repository Kaggle [9]. Firstly, data
description is performed through distribution
representation for each attribute. Results are presented
in Table 2.
Table 2. Data description
Variable
Distribution
CustomerID
Gender
Senior Citizen
Partner
Dependents
Tenure
PhoneService
MultipleLines
InternetService
OnlineSecurity
OnlineBackup
DeviceProtection
TechSupport
This dataset is used in the modeling phase. The
decision tree is applied to the dataset, as a machine
learning algorithm for the development of predictive
models. A decision tree is a well-known algorithm
whose results are easy to understand. Different
learning parameter settings were employed on the
decision tree algorithm to get good models. In the end,
active statistical pruning is used. The reliability of the
active statistical pruning model is 73.46%. The
attribute for which we make the prediction model is
Churn, ie the departure of the client. We ask the
question "Will the client leave or not?". The model we
will choose for further analysis will be the active
statistical pruning-based model since it is the most
accurate and reliable of all models. The accuracy of
such a model is very high because the error of the
model is less than 4% (3.45%) Figure 1 depicts the
model.
Fig. 1: Decision tree model
Sensitivity analysis is performed with the aim to
detect the most important variables for churn
prediction. The results are shown in figure 2.
Fig. 2: Results of sensitivity analysis
The most important attribute is at the root, which is
Contract, then Tenure, and then Internet Service. In
the decision tree, we can see that the probability of the
client leaving is highest when the attribute Contract =
Month-to-month, Tenure <= 14, and InternetService =
Fiber optic is 69.72%. Interpreting these data, we can
conclude that most often customers who renew the
contract period monthly, use the services of the
company for less than 14 months and have fiber
optics.
When developed, a predictive model can be used in
prediction for new, unseen data. One example of
prediction is given in figure 3.
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2023.22.8
Dijana Oreski
E-ISSN: 2224-2678
71
Volume 22, 2023
Fig. 3: Deployment phase: the model used in
prediction for new clients
The decision tree model can be transferred into rules,
which are easy to understand. Figure 4 gives the
extraction of one such rule.
Fig. 4: Extracted rule
5 Conclusion
In this paper, we have demonstrated the application of
machine learning in social sciences in the case of a
decision tree algorithm for the prediction of churn.
The results show the churn prediction could be
accurate using machine learning. This analysis has
shown that the proper machine learning application on
churn data can be efficiently used for the vital
extraction of valuable hidden knowledge from the vast
amount of data generated on daily basis. Future
studies should evaluate other machine learning
algorithms, such as artificial neural networks,
k-nearest neighbors, Bayesian approaches, support
vector machines, and ensemble methods to assess
whether performance is improved.
References:
[1] Shinno, H., Yoshioka, H., Marpaung, S., & Hachiga,
S. (2006). Quantitative SWOT analysis on global
competitiveness of machine tool industry. Journal of
engineering design, 17(03), 251-258.
[2] Kurttila, M., Pesonen, M., Kangas, J., & Kajanus, M.
(2000). Utilizing the analytic hierarchy process (AHP)
in SWOT analysis—a hybrid method and its
application to a forest-certification case. Forest policy
and economics, 1(1), 41-52.
[3] Osuna, E. E., & Aranda, A. (2007). Combining SWOT
and AHP techniques for strategic planning. Economic
Journal. Instituto de Estudios Superiores de
Administración (IESA) Avenida IESA, San
Bernardino, Caracas–Venezuela.
[4] Jeon, Y., & Kim, J. (2011). An application of
SWOT-AHP to develop a strategic planning for a
tourist destination.
[5] Görener, A., Toker, K., & Ulucay, K. (2012).
Application of combined SWOT and AHP: a case
study for a manufacturing firm. Procedia-social and
behavioral sciences, 58, 1525-1534.
[6] Houben, G., Lenie, K., & Vanhoof, K. (1999). A
knowledge-based SWOT-analysis system as an
instrument for strategic planning in small and medium
sized enterprises. Decision support systems, 26(2),
125-135.
[7] Milano, M., O’Sullivan, B., & Gavanelli, M. (2014).
Sustainable policy making: A strategic challenge for
artificial intelligence. ai Magazine, 35(3), 22-35.
[8] Athey, S. (2017). Beyond prediction: Using big data
for policy problems. Science, 355(6324), 483-485.
[9] Kaggle (2022), available at:
https://www.kaggle.com/blastchar/telco-customer-chu
rn
Sources of Funding for Research Presented in a
Scientific Article or Scientific Article Itself
This work has been supported in part by Croatian
Science Foundation under the project
UIP-2020-02-6312.
Creative Commons Attribution License 4.0
(Attribution 4.0 International, CC BY 4.0)
This article is published under the terms of the
Creative Commons Attribution License 4.0
https://creativecommons.org/licenses/by/4.0/deed.en_
US
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2023.22.8
Dijana Oreski
E-ISSN: 2224-2678
72
Volume 22, 2023
Contribution of Individual Authors to the
Creation of a Scientific Article (Ghostwriting
Policy)
The authors equally contributed in the present
research, at all stages from the formulation of the
problem to the final findings and solution.
Conflict of Interest
The authors have no conflicts of interest to declare
that are relevant to the content of this article.