Vicario & Coleman, 2019

A Review of Data Science in Business and Industry

and a Future View
Grazia Vicario1,*,+ Shirley Coleman2
1
Department of Mathematical Sciences, Politecnico di Torino. C.so Duca degli Abruzzi 24, 10129
Torino (Italy)
2
School of Mathematics, Statistics and Physics, Newcastle University, NE1 7RU, UK

Abstract

The aim of this paper is to frame Data Science, a fashion and emerging topic nowadays in the context
of business and industry. We open with a discussion about the origin of Data Science, and its
requirement for a challenging mix of capability in data analytics, information technology and
business know-how. The mission of Data Science is to provide new or revised computational theory
able to extract useful information from the massive volumes of data collected at an accelerating
pace. In fact, besides the traditional measurements, digital data obtained from images, text, audio,
sensors, etc complement the survey. Then we review the different and most popular methodologies
amongst the practitioners of Data Science research and applications. And since the emerging field
requires personnel with new competences, we attempt to describe the Data Scientist profile, one of
the sexiest jobs of the 21st Century according to Davenport and Patil. Most people are aware of the
need to embrace Data Science, but they feel intimidated that they don’t understand it and they
worry that their jobs will disappear. We want to encourage them: Data Science is more likely to add
value to jobs and enrich the lives of working people by helping them make better, more informed
business decisions. We conclude the paper by presenting examples of Data Science in action in
business and industry, to demonstrate the collection of specialist skills that must come together for
this new science to be effective.

Keywords: Knowledge Discovery, Data Scientist profile, Business improvement, Industry 4.0,
SME.

1. Introduction and Rationale of a Data Science review

The paper has been motivated by the need to frame the research area of Data Science in the context of
business and industry. In the last three decades, Data Science has grown up and expanded its research
area attracting interests and researchers from many neighboring scientific fields. This increasing
interest in Data Science arises both from public and private organizations. For example, an internet
shop can exhibit custom-made products and focused advertisements if it is possible to interpret the
data of the customers’ web surfing. Or it can forecast demand and improve its logistics management
if the sold-out data are properly analyzed.
1
* Correspondence to: G. Vicario, Politecnico di Torino, Department of Mathematical Sciences, Corso Duca degli Abruzzi,
24-10129 Turin, Italy.
+ grazia.vicario@polito.it
1
 Vicario & Coleman, 2019

In the area of healthcare, where therapies and diagnoses are more and more digitized and
registered, the use of Data Science methodologies can prevent faulty diagnoses, better detect what
are the most appropriate care plans for the patients, and improve the quality of treatments. Also, in
industrial production, data coming from different work phases are of primary importance to
improve product quality and raise awareness of failures, speed and performance.
Nowadays, the decision processes of organizations are increasingly data-driven. Sensor and
processing tools are accessible to small and medium enterprises (SMEs) thanks to the availability of
open source software. Big Data support is of fundamental importance to face the competition. The
future offers rich possibilities for ever increasing ways of understanding reality, whether it be
through increased integration of linked data sets as in Industry 4.0 (1) and the new government
drives to link administrative data sets, or through exploring hitherto uncharted data sets (2).
A US survey carried out by KPMG2 and based on a sample of 400 CEOs highlighted that
approximately 77% of them harbored mistrust about the quality of the data upon which their
decisions are based. If this were true, all the efforts in analyzing the data would be useless. One
problem may be poor communication between those who present the data and those who read about
it. People belonging to the data analysis team rarely discuss with the decision makers and this can
create an interruption in the communication and development chain. The company realizes the full
benefit of Data Science only if the Data Scientists flank the Managers alongside the decision-
making process, helping them to understand how the data have been processed. On the other hand,
it is also worthwhile for Managers to flank the Data Scientists to reflect the business context in
order to optimize the analysis by having shared goals.
In order to help researchers who want to approach the themes and challenges of Data Science
and to help people who want to pursue a career as a Data Scientist, we discuss in Section 2 the
origin and development of Data Science. In Sections 3, 4 and 5 we review the methodologies,
personalities and applications of Data Science, respectively. We provide some examples of Data
Science projects in Section 6. Final comments and a future view are given in Section 7.

2. Data Science: origin and development

The continuous data flow produced by the internet and by any sensor attached to modern equipment
delivers a huge amount of data. If companies are not able to manage and process the data to their
advantage, they will be outperformed by competitors that are. This is the scenario that gave rise to
Data Science and still supports its development which has been as rapid as the explosive
technological change. Data Science is a science offering methodologies for processing and
interpreting massive volumes of data collected by an increasing number of new devices. These
analytical tasks are difficult to accomplish using just the long-established statistical methodologies.
This science is defined today as an interdisciplinary field, including mathematical methods,
statistics, algorithm developments, qualitative analysis, computer science and, not less importantly,
a practical approach, intending to extract useful information from data, either structured in terms of
quantitative information in a set format or unstructured such as reports, visuals and sounds.
Originally, the denomination Data Science was used by the Turing Award’s winner Peter Naur in
1960, as a synonym of computer science. Later in 1974, Naur used the term Data Science for
referring to data processing methods in their full range of applications (3). But it was in 1996 that
the term appeared in a public writing for the very first time: the International Federation of

2
KPMG is a professional service company and one of the Big Four auditors, along with Deloitte, Ernst & Young (EY),
and PricewaterhouseCoopers (PwC). From https://en.wikipedia.org/wiki/KPMG.
2
 Vicario & Coleman, 2019

Classification Societies organized a meeting in Kobe (Japan) for their biennial conference and
named it “Data Science, Classification and Related Methods”.
Since then, the international community has adopted the name Data Science to indicate an
interdisciplinary field. Nevertheless, Data Science has several times been the focus of different
debates with the purpose of defining its distinction (or its standardization) with respect to Statistics
(4). Just to mention one instance, C. F. Jeff Wu, during his inaugural lecture for the H. C. Carver
Professorship in Statistics at the University of Michigan in 1997 (5), claimed that Statistics should
be renamed Data Science and Statisticians Data Scientists. The new modern methodologies,
however, are pooling the two disciplines of statistics and computer science as in the interaction of
computational algorithms with cognitive science in artificial intelligence and the viewpoint of
machine learning as a marriage of statistics and knowledge representation (6).
Leo Breiman, Statistician at the University of Berkeley, has been of a different opinion. In 2001
he said:
There are two cultures in the use of statistical modelling to reach conclusions from
data. One assumes that the data are generated by a given stochastic data model. The
other uses algorithmic models and treats the data mechanism as unknown. The
statistical community has been committed to the almost exclusive use of data models.
This commitment has led to irrelevant theory, questionable conclusions, and has kept
statisticians from working on a large range of interesting current problems. Algorithmic
modelling, both in theory and practice, has developed rapidly in fields outside statistics.
It can be used both on large complex data sets and as a more accurate and informative
alternative to data modelling on smaller data sets. If our goal as a field is to use data to
solve problems, then we need to move away from exclusive dependence on data models
and adopt a more diverse set of tools. (7)
Breiman’s attitude is reminiscent of a famous phrase by George Box who said:
All models are wrong, but some are useful. (8).
This distinction between Statistics and Data Science is now accepted worldwide, also in university
curricula, where classical statistical analysis is taught in courses based on Statistics and new courses
on Data Science aim to give a more widely addressed education including algorithms and machine
learning.
William S. Cleveland introduced Data Science as an independent discipline in 2001 (9). He
considered that six technical areas encompass the field of Data Science: multidisciplinary
investigations, models and methods for data, computing with data, pedagogy, tool evaluation, and
theory. Subsequently in 2002, the International Council for Science: Committee on Data for Science
and Technology (CODATA) founded the first journal on Data Science
(https://datascience.codata.org/). Immediately thereafter, Columbia University started the
publication of the Journal of Data Science (http://www.jds-online.com/), whose scope is providing
a platform for all data workers to present their view and exchange their ideas”.
Therefore, recently a literature on Data Science has developed, contributing to the information
exchange and creating communities joined by the fascination for data, data analysis and its
applications. Just to cite one for all, the finance and financial technology community adopts a
modern terminology: Fintech Data Science. In fact, one of its representatives, Giudici (10), puts
forward the idea of using the terminology Data ScienceS, because Data Science is an integrated
process of activities (definition of the objectives of the analysis, selection and processing of the data
to be analyzed, statistical modelling and interpretation, implementation and evaluation of the
obtained statistical measures), and the different knowledge domains possibly have different
objectives expressed in different languages.
3
 Vicario & Coleman, 2019

Moreover, since the beginning of the 21 st century, Data Science has grown in different sub-
subjects; among the most important we can mention Knowledge Discovery in Databases, Data
Mining, Artificial Intelligence, Machine Learning and Deep Learning, setting the tone for the
development of other related areas. In Europe, the European Association for Data Science was
founded in 2013, playing an important role in spreading and making more and more popular the use
of the terms Data Scientists and Data Analyst, in Europe and especially in the business field.
Summarizing these concepts and opinions, the aim of Data Science is to clean, prepare and
analyze different data sets extracting meaning from data, thus being related closely to Statistics;
nevertheless, Statistics is used as an instrument by Data Science to reach its purpose, flanked by
information technology, in particular programming for big data analysis, and using a distinctly
practical approach and a decision making implementation.

3. Data Science: Knowledge Discovery in Data Bases

Since the early 40’s, a number of new terms have come into common use, thanks to Information
Technology. Thus, we need to distinguish between data (basic element, usually made of symbols),
information (outcome of processing data aimed at organizing, interpreting and contextualizing data
for acquiring meaning) and knowledge (set of organized and processed information for the purpose
of spreading experience, understanding and competences related to practical problems in industry
and business). We could imagine a sort of chain of acquaintance that moves from data to
knowledge, with the final objective of deciding upon actions, behaviors and decisions. Therefore,
the transition from data to knowledge is strategically important and this is the Data Science
proposition.
To fulfil this objective, Data Science takes advantage of methodologies that are part of disciplines
such as Mathematics, Statistics and Computer Science for analyzing and mining data to achieve
knowledge. Its impact is so decisive that it is considered as the fourth paradigm of science, namely
besides science being empirical, theoretical and computational it is also data-driven. That’s why
Data Science may be considered as a member of a larger family of research fields, involving both
academic researchers and private companies, namely Knowledge Discovery in Databases (KDD).
KDD refers to the process that leads to extracting high-level knowledge from low-level data.
Nowadays, it is one of the most attractive, fascinating and ever-evolving fields to work in, thanks to
continuous improvement of the data warehouse and ever-growing use of Big Data3 and
consequently of Big Data Analytics (aiming at transforming the huge, heterogeneous and persistent
amount of data into usable information).
Big Data originates from ever more innovative multimedia and is characterised by being so big that
traditional statistical methodologies are no longer usable. Maybe it is for this reason that a group of
enthusiasts of Big Data claims that such abundancy of data means that its analysis does not require
a theoretical basis. In 2008, Chris Andersen (11), published a provocative paper entitled: The End of
Theory: the Data Deluge that Makes the Scientific Method Obsolete. He states:
“All models are wrong, but some are useful”. So proclaimed statistician George Box 30
years ago, and he was right. But what choice did we have? Only models, from
cosmological equations to theories of human behavior, seemed to be able to
consistently, if imperfectly, explain the world around us. Until now. Today companies
like Google, which have grown up in an era of massively abundant data, don't have to
settle for wrong models. Indeed, they don't have to settle for models at all. ….. Scientists
are trained to recognize that correlation is not causation, that no conclusions should be
3
The name Big Data refers to huge amounts of data that require ad hoc procedures to be captured and analyzed. In 2013,
the term Big Data entered the prestigious Oxford English Dictionary, going against the editing rules that a new word
needs to be in use for at least 10 years.
4
 Vicario & Coleman, 2019

drawn simply on the basis of correlation between X and Y (it could just be a
coincidence). Instead, you must understand the underlying mechanisms that connect the
two. Once you have a model, you can connect the data sets with confidence. Data
without a model is just noise. But faced with massive data, this approach to science —
hypothesize, model, test — is becoming obsolete.
This thinking started off a vibrant debate that is still alive today in the scientific community. In this
paper, we do not want to take part actively in this debate either in favour or against the Anderson
mindset but we cannot miss the opportunity to express our opinion concerning statistical models.
The useful model cited by Box is linked to a physical or simulated experiment with the aim of
prediction; it is not required to reproduce the complexity of the phenomenon, but only to give a
significant support to the prediction algorithm. Our conviction is that there is an urgent need for a
new generation of computational theories and tools to assist humans in extracting useful
information (knowledge) from the rapidly growing volumes of digital data as stated by Usama
Fayyad et al. (12) and other Data Science practitioners.
Most of the work is still handled by humans, who need to make the correct decisions so as not to
spoil the data and, most importantly, to achieve an adequate data significance. Appropriate a-priori
knowledge of the problem is needed to correctly handle data research, data preparation, selection
and cleaning, to be sure not to use inappropriate data, as well as to properly interpret the results of
mining, in order not to misunderstand the information extracted by the data analysis. The KDD
process in data bases is essential, even if it is very complex. It may require several iterations but it
does lead to success in data research problems if human participation and interaction are effective.
According to Usama Fayyad et al. who outlines the basic and practical steps of the KDD process
suggested in (13), the steps are:
1. Understand the application domain, gain an adequate a-priori knowledge and identify correctly
the goal of the KDD process from the customer’s viewpoint;
2. Select or create a target data set on which the discovery is to be performed;
3. Clean and pre-process data: remove possible noise, by collecting enough information about how
to identify it; decide strategies about missing data or missing data fields; account for time-sequence
information and known changes;
4. Find useful features to represent the data depending on the objective of the task, trying to reduce
dimensions or find invariant representations for the data to lighten the research;
5. Find a data mining method based on statistics that conforms to the task of the Step 1
(summarization, classification, regression, clustering, etc.);
6. Perform an exploratory analysis, a model and hypothesis selection, choosing the algorithm(s) and
selecting method(s) for searching for data patterns;
7. Do real analysis using data mining methods, (including classification rules or trees, regression,
and clustering);

8. Explain the mind pattern, maybe including visualization of data to extend some useful visual
information;
9. Act on the discovered knowledge: use it directly, incorporate it into another system for further
action, document and report it. Check for and resolve potential conflicts with previous beliefs.
After concluding the initial steps, in which the necessary data has been found, the Data Mining
process is the central core applicable to different purposes, for example verification of hypotheses
5
 Vicario & Coleman, 2019

made by the user, discovery of new patterns never seen before or description, if patterns have to be
represented in a human-understandable form. Most Data Mining methods are based on tried and
tested techniques belonging to Machine Learning, Pattern Recognition or Statistics.
The Data Mining denomination is frequently associated with Data Science, sometimes even
confused with it, because of the large amounts of mathematical and computer science methods used
by all of these disciplines. However, they are quite different, because the former is a process for
discovering patterns in large data sets (with previous knowledge on it), extracting information and
transforming it into an understandable structure for further use, typical for appropriate decisions;
whereas in contrast, Data Science is a larger family inclusive, for example, of data cleansing,
preparation and final analysis on data sets of every type.
Finally, we consider that the mainstream of Data Science is Data-Driven Decision Making, namely
the objective to take decisions based on accurate and thorough data analysis rather than on
guesswork and intuition.

4. The Data Scientist

After having framed the concept of Data Science, the question arises: who is or, what is expected to
be the Data Scientist? We try to provide a Data Scientist profile. The Data Scientist is the person
who understands how real problems relate to available data and so can make the best use of data for
creating added value. The Data Scientist manages data, guarantees its integrity and accessibility and
mines useful information from data to provide knowledge and prediction, and support decision
making.
And what are the competences that a Data Scientist is expected to have? First of all, the Data
Scientist has to be able to analyze problems in a logical way and with understanding of the real
underlying business or practical issues. They have to be able to deal with a large number of
techniques. If you Google “data scientist job description”, you can find many web sites listing these
skills (about 144,000,000 results (0.48 seconds): Google informs us) and other links for searches
related to the Data Scientist job description. The general agreement is that the following statistical
methods should be available to everyone considered to be a Data Scientist:
 High dimensional geometry - nowadays, vectors with many components are habitually used
for profitably representing situations in data, even if the vector representation is not the
natural original choice in collecting the data.
 Singular Value matrix Decomposition, Principal Component Analysis and of course Matrix
Algebra should be familiar tools.
 Modeling data with a mixture of Gaussians, a standard tool in Statistics; clustering that
refers to partitioning a set of elements into subsets according to a criterion or just identifying
existing natural clusters.
 Random Graphs (they are commonly used in many of today’s contexts, such as World Wide
Web, Internet, social networks, journal citations,…), Random Trees and Random Walks on
Directed Graphs, Markov Chains and Neural Networks.
In addition, competence in Machine Learning (ML) is surely desirable and advantageous for the
Data Scientist. ML is an interdisciplinary field that unifies applied statistics and computer science,
aiming to estimate complicated functions by using some algorithms that can learn from previous
data. Thus, ML algorithms are able to adapt to the problem they are analyzing. They are used for
tasks that are too difficult to be solved with fixed programs: the algorithm is not created to do
something, but it is created to learn to do something. They have to process vectors with large
amounts of data, both in the training period and in the test period, concepts that come with Big
Data. Analytical models are usually developed on a training data set, tested on a test data set and
6
 Vicario & Coleman, 2019

verified on a further data set before being applied to the whole data set to ensure applicability to the
full range of data. This concept is less common in traditional statistical analysis where data are
often generated for specific cases and may be in short supply, and models are more likely to be
assessed by using leave-one-out techniques.
Algorithm accuracy is measured by the Task the algorithm is created for (different types of tasks
are: Classification, Classification with missing inputs, Regression, Transcription, Anomaly
Detection, De-noising), by the Performance the algorithm has on that task (it can vary from case to
case, for example is it better to have an output with few important mistakes or one with several
medium mistakes?) and by the Experience it is allowed to have during the learning process
(quantity of training sets the algorithm is allowed to analyze before starting the test set or the real
analysis). Of course, the less data needed to get an adequate analysis, the better the algorithm is
considered to be.
Just to pinpoint the required competences of the Data Scientist, we mention the distinction
between unsupervised and supervised learning algorithms. Unsupervised learning algorithms
experience a data set containing features and try to learn useful properties of the structure of the
data set, such as the probability distribution from which the data set was generated or a method to
de-noise the whole data set. Basically, the algorithm experiences several examples of vector 𝒙 and
attempts to learn the associated probability distribution or some properties of it. Supervised learning
algorithms experience a data set containing features in which each example of a random vector X is
associated with a label or a target of an associated value or vector Y; then the algorithm tries to
predict Y from X, usually using the underlying probability distribution. All in all, we have to admit
that there is no neat distinction between unsupervised and supervised learning algorithms, because
in theory supervised algorithms ought to be guided by the label, while unsupervised should not; but
the same algorithms can often be used in both cases, with minimal differences.
There are some problems, especially with the Artificial Intelligence tasks, that ML cannot solve,
for various reasons. Deep Learning (DL), or Deep Neural Networks, is a new autonomous
discipline created to overcome the limits of classical ML algorithms; DL derives from ML, but it
may be considered part of ML itself. Nowadays, there are a lot of different types of DL algorithms,
developed to complete specific tasks, and that obviously adapt to the one they have been created for
(object recognition, speech recognition, bioinformatics, and others). DL algorithms may have
different architectures, but they refer to training many-layered networks of nonlinear computational
units such as Neural Networks do (input layers, output layers or hidden layers, deep feedforward
neural networks, …. are terms that DL and Neural Networks share). There have been many
outstanding successes in DL applications. For example, the ImageNet challenge involves
recognizing 15 million images; network analysis with over 150 layers succeeded in making fewer
than 5% errors in 2016 thereby beating performance achieved by human beings (14).
For a comprehensive examination of how Statistics has evolved in the last decades, we suggest
seeing the recent book of Efron and Hastie (15). The authors discuss the most influential and
popular topics in Statistics and how they have been reshaped by modern computing 4. It is evident
that the Data Scientist is not expected to be a Computer Scientist, but has to have a fair degree of
familiarity with Information Technology. Summarizing, the Data Scientist does have cross
competences in Statistics, Mathematics and Computer Science. Since we live in a digital epoch
where anything may be a source of digital data, the professional who is able to manage data and to
extract value from it will be the most coveted in the working environment. Not by chance did
Davenport and Patil (16) claim:

4
Bradley Efron was awarded the International Prize in Statistics and he will receive it during the next ISI (International Statistical Institute) World Statistics
Congress to be held in a Kuala Lumpur, August 2019 (http://www.isi2019.org/). He is a professor of Statistics and Biomedical Data Science at Stanford
University. The recognition is mainly due to the "bootstrap" method he developed in 1977 for assessing the uncertainty in scientific research. The method had
an extraordinary impact across many experimental scientific fields.

7
 Vicario & Coleman, 2019

The Data Scientist: the sexiest Job of the 21st Century.

The demand for Data Scientists has led Universities to rush to offer new curricula aimed at
training Data Scientists. Most of the courses offered are at post graduate level, although
undergraduate courses are also emerging. They vary in emphasis and content but the most common,
relevant features are solid theoretical foundations in Statistics and Computer Science combined with
practical experience of diverse applications. There must be a good two-way exchange between
academia and applications whether interdisciplinary within the university or with business and
industrial partners, (17). Universities are ideally placed for promoting a valuable interaction between
business and industry by including the supervision of postgraduate projects and placements as part of
Data Science training.
Data Science Master’s degrees are of interest to all University Faculties because of the potential
of attracting students and for collaborative projects. To meet the growing interest from students and
the need from employers, many courses have been fast-tracked into existence. A nice method of
researching the necessary content, need and practicalities was described by Brown (18) at the
University of Canterbury, New Zealand. Members of all faculties were invited to a 2-hour session
and asked to write down what they considered to be the characteristics of a Data Scientist who
successfully completed their prospective Master’s degree in Applied Data Science. They then wrote
down the characteristics of the likely applicants to their Master’s degree and worked out what
modules and activities were needed to get from one to the other. The result was a 12 month taught
degree with modules including data analytics and digital humanities, and an industry project. The
focus was on flexibility and on creativity and it has seen a burgeoning number of applicants and
successful graduates.
We conclude reporting a statement of James Stephen Marron, Professor at UNC-CH Department
of Biostatistics, University of North Carolina:
I think it’s time for Data Science to consider the concept of team work. Data Science
problems with one person working alone are mostly solved, but for solving big challenges it is
necessary to have the work of a team of data scientists with different skillsets.
It is certainly true that the growing importance of Data Science and its increasing application in
many areas, including SMEs, does mean that it touches upon many different fields of expertise.
Where data science is firmly embedded, such as in some government services in the UK, there are
sub-divisions within the Data Science community containing specialist experts in one or more of the
main subject areas. The requirement of being a T shaped person, who has wide knowledge base as
well as deep skills, is common in many work areas and is also an asset in everyday life; but in the
Data Science context, we consider it to be essential.

5. Development of Data Science in Business and Industry

The development of Data Science has been motivated by the explosion of data in the digital age. It is
worth noting that many of the mathematical, statistical and machine learning techniques in Data Science
have been around for many years. What has changed is the availability of massive amounts of data that
are now stored rather than being merely observed and then overwritten, and more recently the
realization that profound insights and business advantage can be gained from analyzing this data.
Data Science as a profession is increasingly being seen as high paid, glamorous and much sought after.
This changing tide is encapsulated in the article in the New Yorker shown in Figure 1.

8
 Vicario & Coleman, 2019

Figure 1 The changing tide in professions

Previously it was the creative “Madison Avenue” people who contrived to sell in the most effective
way but now behavioral profiling and customer segmentation techniques employed by mathematicians
and statisticians are proving that scientifically targeted advertising is much more profitable than even
the most elegant and subtle advertising campaigns. New digital displays and fast interactive processing
of customer profiles enable promotions to be presented to people just when they are likely to be most
susceptible. Hence the math men are overthrowing the mad men.
Data Scientists are now seen as useful and desirable. There has been a lively debate about the
relationship between Statistics and Data Science (19). Statistics plays a key role in Data Science and it is
worthwhile considering how statisticians should manage that role. Many data analytical practices
revolve around opaque solutions in which data is fed into a black box, calculations are made and an
answer comes out. Black box techniques mask the algorithms being used and just report an answer. It is
true that there can be good reasons for encasing the methodology in a black box, for example if it is
highly complex or so that it cannot be interfered with. However, the downside is that black boxes
alienate statisticians and encourage a cavalier attitude to statistical detail and theoretical niceties that is
frustrating to statisticians and maybe damaging in the long term.
Black box data analytical solutions for prediction lack robustness against changes in influential
variables within the data sets. The solutions are often based on assemblies of models and predictions are
averaged out using a range of methods, not all of which are appropriate for the type of data being input.
It is not easy to find out which predictors have had the major influence on the prediction. One predictor
may be much easier to collect than others and have higher quality, but the advantage of using this
variable may not be realized unless further checks are made on the solutions to see the effect of
swapping variables. Black box users are not forced to check their data before analysis and there is often
little emphasis on residual analysis so that variables with gross outliers are not detected. Users may miss
data errors and opportunities for finding key subsets and obvious explanations for apparent patterns. But
the black box user tends not to mind any of this provided their prediction is better than the one before
and seems to work in the short term. In summary, the problems with dependence on black boxes are that
 Algorithms, tuning parameters, subtle effects and assemblies are not accessible;
 Black boxes may be adequate for short term solutions but robustness to change is not certain;
 Skills to understand the analytical methods are not valued and developed.
Business people tend to like black box approaches as they are superficially easier to understand.
Statisticians need to reclaim the field rather than let core black box services take over; they need to
stand up for the importance of checking models and not just believing the solutions offered. Statisticians
need to proclaim the many important roles for which they are needed, such as: to check quality of input
data, evaluate costs of obtaining variables and using proxies, conduct sensitivity experiments, construct
and validate models.
Companies have a love-hate relationship with Data Science based on primordial fear of numbers and
a strong desire for the benefits that analysis can bring (20). We need to encourage staff by emphasizing
9
 Vicario & Coleman, 2019

that there are many positive outcomes from Data Science, such as less waste of time and effort, more
time to devote to their main core work and greater profitability and security in employment.
The growth of Data Science affects academia, business and industry; it has a positive influence because
it focuses attention on statistical techniques and raises awareness of data in all walks of life in the same
way as did Six Sigma, Total Quality Management and other business improvement initiatives (21). Data
Scientists need to be numerate and the growth of Data Science encourages enthusiasm for education in
all STEM (Science, Technology, Engineering, Mathematics) subjects at all levels. As well as
undergraduate and postgraduate degrees, there are also increasing opportunities for vocational training.
The growth in Data Science therefore gives a welcome boost to interest in Mathematics and Statistics
from high school to tertiary education and in continuing professional development. An example of
government interest in this area is the Knowledge Transfer Partnerships funded by Innovate UK. These
provide up to 2/3 of the costs of one to three year projects in which a postgraduate research associate is
embedded in a company to develop new expertise, supervised by a university academic. These projects
are an effective way for SMEs to grow their business with state-of-the-art guidance but less risk and
financial outlay (22).
In some senses, we are in a golden era with great opportunities afforded by the expansion of Data
Science before data analytics becomes so entrenched that there are fewer creative opportunities for
developing bespoke solutions. We need to ensure that the quality and purity of Statistics is maintained
not only for our own professional sensibilities but more importantly because it is only when Statistics is
applied sensitively and correctly that the optimum benefits of Data Science are achieved.

6. Examples of Data Science in Action

To demonstrate the collection of specialist skills that must come together for Data Science to be
effective, we consider some examples. In each case study the business need is clearly stated, the data
integration is explained, and then the statistical analysis and the outcomes are described.

Automotive retail sector

Automotive after sales is a business that generates a massive quantity of data. Each day, vehicle owners
and service engineers consult catalogues to find the right parts to carry out repairs to their vehicles. In
this case study the business partner is a small to medium enterprise (SME) handling big data from
catalogue look-ups and other data associated with the automotive after sales market sector. The business
motivation is to increase their service offering to customers focusing on some issues that were identified
by the customers themselves and on other opportunities that became apparent after applying exploratory
statistical analysis and data visualization of the data. Empirical data analysis can be used to explore
many different scenarios. Applying IT skills to amalgamate diverse sources of information followed by
statistical analysis to identify patterns can bring valuable business insight. In this way, the Data Scientist
is helping to improve the SME’s current business and also potentially identifying new products and
revenue streams (23). We consider three specific examples relating to: mileage; return rates and original
equipment manufacturing.
Firstly, looking at the mileage of vehicles coming to a garage to have a repair carried out, the
massive quantity of data available highlights some distinct differences in different types of vehicles.
Data on the vehicle and the parts fitted are integrated and checked in preparation for analysis. Data
dimensions include make and model of vehicle, type of repair, age of vehicle, date of visit to garage,
mileage at time of repair and parts fitted. The data were analyzed by constructing empirical cumulative
distribution curves for mileage for specific makes and models of vehicle and type of repair. Figure 2
compares the curves of mileage in vehicles coming to the garage for brake disk replacement for 3
different popular saloon cars. The plot shows that cars of type A go to the garage with much lower
mileage than for either of B or C. In this data set 50% of cars of type A have mileage 65,000 or less
whereas for cars of types B and C the percentage is nearer 25%. This implies that cars of type A have
brake disc replacement earlier and are in this sense a less desirable car.

10
 Vicario & Coleman, 2019

65000
Variable
100 Car A
Car B
Car C

80

60
Percentage

50

40

25
20

0 50000 100000 150000 200000

Mileage

Figure 2 Brake Disc replacement for 3 popular saloon cars.

The outcome of the Data Science analysis is an empirical tool that can be updated as more data arise
and can be used by the garage to determine necessary stock levels, likely co-morbidity when vehicles
come in for a particular repair and a service to customers when considering buying a used vehicle.
Secondly, return rates for specific items bought from a catalogue are shown to differ widely; those items
that exceed control limits in a funnel plot of return rates merit further investigation which may include
checking how the parts are presented in the catalogue.
Thirdly, only the original manufacturer knows exactly which original equipment was fitted to a
specific vehicle and generally their parts are more expensive than the generic copies which are made by
other companies. There is always some uncertainty when buyers are choosing between alternative
products and this opens the way to some business opportunities using data analytics. Suppliers of a
particular spare part for a vehicle were interested to know if their part would fit vehicles other than the
one for which it was designed. A search through the database of part dimensions and a matching process
led to identification of several other vehicles that could also use the part (24).
This case study looks at adding value to administrative and operational data from a range of company
sources and demonstrates where improvements can be made in 3 areas: in the service offering by
garages; in catalogue clarity and usability; and in reach for suppliers.

Shipping sector
The shipping industry is the nucleus of global trade and is highly sensitive to fuel prices. Fuel costs
represent over 50% of the total operating costs of a vessel (25) and contribute a significant portion of the
total transportation cost of cargo. Fluctuations in the price of crude oil and stricter environmental
regulations on the emission of noxious and greenhouse gases are influential factors in the operation of
the shipping industry. Data Science has been used to develop new products offered by a small to
medium enterprise (SME) dealing with shipping performance data.
Sensors attached to ship’s engines record fuel consumption which is displayed in real time on the
ship’s bridge. As well as providing a check on fuel theft and gross errors in engine function, this data

11
 Vicario & Coleman, 2019

can be used for many other purposes. The business motivation of this case study is to provide a decision
support tool to aid the scheduling of ferry services.
Fuel consumption is related to the ship’s speed over ground. The Company operational data is
especially useful when enhanced with open data on weather and tides. The components of weather that
most affect fuel consumption have been shown to be head wind and cross wind. Tidal data is important
in certain shipping routes and in this case study, the tides around the UK are quantified using algorithms
and data on freely available websites.
Company data on fuel consumption over the whole journey of a ferry carrying out a daily return
service on a particular route is integrated with weather and tide data after adjusting them for location
and time granulation issues. The data are analyzed using multivariate statistical analysis to produce a
predictive model for fuel consumption based on known tides and predicted weather.

7 7 7 7 7 7
01 01 01 01 01 01
4-2 4-2 4-2 5-2 5-2 5-2
-0 -0 -0 -0 -0 -0
10 12 14 08 10 13
Outward Return Variable
25.0 FC
Predicted
UCL
Fuel Consumption

22.5 LCL

20.0

17.5

15.0

17 17 17 17 17 17
20 20 20 20 20 20
04- 04- 04- 05- 05- 05-
- - - - - -
10 12 14 08 10 13

Date

Figure 3 SPC chart for fuel consumption. UCL and LCL are upper and lower control limits. Lines for
Predicted and actual fuel consumption (FC) are shown.

Figure 3 shows a statistical process control chart constructed using the residual fuel consumption
after fitting a regression model. It can be seen that during the time shown in the control chart, the
outward journey generally requires less fuel than the return journey due to the tide. Journeys that have
fuel consumption within the upper and lower control limits allowing for the specific weather and tide
can be considered as satisfactory. Journeys exceeding either the upper or lower control limit merit
further investigation as to possible causes for the fuel used. In this case, all journeys were within the
control limits. Using control limits avoids the shipping company unnecessarily spending time checking
when there is no real change in the performance of the vessel.
Data Science enables the shipping company to understand the performance of its ships and the
variation in fuel consumption costs and corresponding emissions due to weather and tides. Where
sufficient flexibility exists, journeys can be scheduled to minimize cost. In this way the analysis
provides a valuable management decision making tool based on data already collected and available.
In another application of Data Science in the SME, the most economical speed for the vessel to travel
is estimated. Ships are subject to dock trials while the ship is being constructed, builder’s trials when the
ship is completed and sea trials when the ship starts in service. These trials indicate the economical

12
 Vicario & Coleman, 2019

speed for travel. However, over time, the ship’s performance changes due to fouling, damage caused by
minor collisions, etc. Employing statistical thinking to design trials for the ship during service leads to a
reassessment of the most economical speed and consequent improvement in the business.
This example looks at adding value to fuel consumption data collected for the purpose of monitoring,
and developing new products for the SME thus increasing sales and widening their customer base. The
work involves detailed knowledge of the shipping scenario, handling and amalgamation of vast sets of
fast moving data and access to open data, statistical analysis and then business know-how to convert the
findings into a valuable product for the company to sell. It shows the value of the combination of all 3
aspects of Data Science.

Social housing sector

Social housing accounts for about 50% of rental properties in the UK. It aims to provide affordable
accommodation and is controlled by strict governmental regulations. Certain social housing providers
have tens of thousands of properties, but still rent collection, repairs and maintenance are usually
outsourced to bespoke software providers. The business motivation of this case study is to identify
tenants in danger of falling into arrears before their debts become too difficult to manage. These tenants
can then be assisted to make their payments and the housing provider can potentially reduce the level of
their arrears caseload and account processing time, thus increasing income collection.
In this example, a small to medium enterprise (SME) software house is using Data Science to
maximize the insight that can be obtained from their vast reservoir of interesting data on rent balances,
property repairs and empty properties, or voids (26). Social housing data is complex and includes
information about the property, about the tenant and the payment details during each tenancy. This is
confidential data that has to be handled with absolute adherence to data protection laws. There needs to
be extensive data amalgamation from multiple tables dealing with property characteristics, tenancy
information, ongoing payments and interactions between the tenant and rent officers. Hence, preparing
the data for analysis is a major part of the project.
Weekly arrears data is analyzed using time series and machine learning methods including cluster
analysis. Figure 4 shows the main pattern of rent balance profiles over different periods of time.
Balances are negative when rent has been paid and thus the tenant is in credit, while positive balance
means there is a debt. Cluster analysis identified a major cluster of tenancies and multiple exceptional
clusters. In Figure 4, the left-hand side main cluster illustrates temporal patterns. The top plot shows that
rent arrears follow a downward trend, while the middle plot shows that rent arrears increase throughout
the month and the lower plot shows the peak in winter. The plots on the right-hand side of Figure 4
demonstrate typical outliers, with an overall upward trend over the three-year period in the top plot and
no clear monthly pattern in the middle plot; the lower plot shows a more variable yearly pattern than the
main cluster with a shorter period of being in credit than for the majority.

13
 Vicario & Coleman, 2019

Figure 4 Cluster analysis of rent balance profiles

Statistical and machine learning analysis have been used to predict arrears, to model the weekly,
monthly and yearly patterns in rent balances and to identify clusters of tenants and understand the
important features in each cluster. In addition, data visualization has been shown to be a very powerful
tool for giving social housing providers insight into their business. This example illustrates a Data
Science application in a sector hitherto unfamiliar with data analytics and shows that there are enormous
benefits from using vast routinely collected operational data resources for the purpose of day to day
running of the business, in a sensitive way.

7. Final Comments and Future View

We would like to mention a representative episode from a New York Times article in 2004
(https://www.nytimes.com/2004/11/14/business/yourmoney/what-walmart-knows-about-customers-
habits.html) that is mentioned in Foster Provost F. et al. (27):

Hurricane Frances was on its way, barreling across the Caribbean, threatening a direct
hit on Florida’s Atlantic coast. Residents made for higher ground, but far away, in
Bentonville, Ark., executives at Wal-Mart Stores decided that the situation offered a
great opportunity for one of their newest data-driven weapons … predictive technology.
A week ahead of the storm’s landfall, Linda M. Dillman, Wal-Mart’s chief information
officer, pressed her staff to come up with forecasts based on what had happened when
Hurricane Charley struck several weeks earlier. Backed by the trillions of bytes’ worth
of shopper history that is stored in Wal-Mart’s data warehouse, she felt that the
company could ‘start predicting what’s going to happen, instead of waiting for it to
happen’, as she put it.
The authors wonder why data driven prediction is useful in this situation: for predicting the need
of an increased amount of water bottles? Or for discovering that a particular product was sold out in
those days and this fact may be related to the hurricane? Or trying to learn from the data in previous
hurricanes if there were particular needs? We may go on with more points at issue, but these facts
are not difficult to be predicted or to be explored. It’s surely more interesting to look for more
sophisticated and more aimed models, mining the huge amount of data stored by Wal-Marts before
the hurricane Florence. And so it was: The New York Times article (Hays C.L., What Wal-Mart
Knows About Customers' Habits, November 14th, 2004) continues:

14
 Vicario & Coleman, 2019

… the experts mined the data and found that the stores would indeed need certain
products —and not just the usual flashlights. “We didn’t know in the past that
strawberry Pop-Tarts increase in sales, like seven times their normal sales rate, ahead
of a hurricane,” Ms. Dillman said in a recent interview. “And the pre-hurricane top-
selling item was beer.”
We report the anecdote to highlight the main strength of Data Science: the Data-Driven Decision
Making i.e. the objective of coming to decisions based on rigorous and accurate analysis of data
rather than on intuitive insight alone. It has been shown that data dependency in a company makes it
at least 6% more productive. Business needs in-depth analysis and to be provided with a detailed
overview of the current situation and of the prospective changes.
Unfortunately, as the change in business models by companies is to become more and more data
driven, they come up against the inadequacy of supply of qualified personnel, i.e. Data Scientists:
their number is not sufficient in the face of such a fast and radical change in business. For this
reason, universities are preparing academic and post-academic courses with different but relevant
curricula. This testifies to the good job prospects for Data Scientists and the extent to which the
ability to analyze data should become a key tool for all ambitious managers.
In order to ensure good practice, specialists in the different parts of Data Science need to engage
with each other and embrace their different ways of thinking. It was realized with Six Sigma that a
business focus was vital and this is even more the case with Data Science. The challenges that are
tackled and the solutions found must be tailored to issues of strategic importance to the business.
Otherwise the Data Science becomes more of an academic exercise and companies downplay the
role that Data Science could have in their company thus restricting the opportunities for innovative
applications. Data Scientists need to be creative in their thinking, central to the business and
constantly mindful of the advantages of having an open and free exchange with specialists in
different areas. Good Data Science requires effective communication with other professionals and
for all to act together in harmony.
Statisticians need to be confident that what they have to offer is valuable and important.
Companies need help to understand this new data centric world and one way is to present case
studies which clearly show the benefits of correct analysis and its supremacy over black box
solutions. Reports of successful and less successful Data Science projects need to be readily
available not only in academic publications but also in the trade press with which business people
are familiar.
This requirement is being addressed by a growing literature of case studies and opinions
available in the popular press, for example the article by Shan (28), and in specific issues of
journals, for example the paper by Giudici (10). Text books are slower to follow suit but there are
some excellent publications incorporating statistical techniques in a Data Science context, such as
James et al (29) and the more all-encompassing Data Science Handbook (30). Our future view is
one of optimism for the continued progress of statistical involvement in Data Science and the
increasing importance of Data Science as a modus operandi for companies. The enormous financial
benefits on offer ensure that there is adequate funding for research and new ideas and methods are
emerging all the time.

15
 Vicario & Coleman, 2019

References

1. Coleman, SY. Data science in Industry 4.0. [book auth.] ECMI. ECMI conference, Budapest June 2018, in
ECMI book subseries of Mathematics in Industry. s.l. : Springer, 2019.

2. Ahlemeyer-Stubbe, A and Coleman, SY. Monetising data – how to uplift your business, Wiley. London :
Wiley, 2018.

3. Naur, Peter. “Concise Survey of Computer Methods”. Lund, Sweden: Studentlitteratur. 1974. Retrieved
from: http://www.naur.com/Conc.Surv.html.

4. Joseph, Hugh A. Chipman and V. Roshan. A Conversation with Jeff Wu. Statistical Science. 2016, Vol. 31,
4, pp. 624–636.

5. Wu, C.F.J. “Statistics = Data Science?” 1997.

6. Flach, P. Machine Learning: The Art and Science of Algorithms that Make Sense of Data. s.l. : Cambridge
University Press, 2012.

7. Breiman, L. (2001) Statistical Modelling: The Two Cultures. Statistical Science. 16(3), 199-231. , p. .

8. Box, G. E. P. (1976). Science and Statistics. Journal of the American Statistical Association, 71: 791–799.
doi:10.1080/01621459.1976.10480949.

9. Cleveland, W. S. (2001). Data science: an action plan for expanding the technical areas of the field of
statistics. International Statistical Review / Revue Internationale de Statistique, 21–26.

10. Giudici, P. Financial data science. Statistics & Probability Letters. May 2018, Vol. 36, pp. 160-164.

11. Wired., Chris Andersen (2008) The end of Theory: the Data Deluge makes the Scientific method
Obsolete:.

12. Fayyad, U., Piatetsky-Shapiro, G., Smyth, P. From Data Mining to Knowledge Discovery in Databases.
Al Magazine, 17(3). 1996. Retrieved from:
https://www.aaai.org/ojs/index.php/aimagazine/article/viewFile/1230/1131.

13. Brachman, R., and Anand, T. 1996. The Process of Knowledge Discovery in Databases: A Human-
Centered Approach. In Advances in Knowledge Discovery and Data Mining, 37–58, eds. U. Fayyad, G.
Piatetsky-Shapiro, P. Smyth, and R. Uthurusamy. Menlo Park, Calif.:.

14. The Great Expectations of the ImageNet Challenge. Soft., Science. s.l. :
https://www.scnsoft.com/blog/imagenet-challenge-2017-expectations, 2017.

15. Efron, B and Hastie, T. Computer Age Statistical Inference: Algorithms, Evidence, and Data Science. s.l. :
Cambridge University Press, 2016.

16. Patil, d and Davenport, TH. Getting control of Big Data. Harvard Business Review. 2012.

16
 Vicario & Coleman, 2019

17. Coleman, SY and Kenett, RS (2017) The Information Quality Framework for Evaluating Data Science
Programs. Available at SSRN: https://ssrn.com/abstract=2911557.

18. Brown, J. (2018). Leading change: Developing a new Applied Data Science programme. ICOTS10,
Kyoto, http://iase-web.org/icots/10/proceedings/pdfs/ICOTS10_1H1.pdf?1531364187.

19. kdnuggets discussion. [Online] 2013. https://www.kdnuggets.com/2013/04/data-science-end-

statistics-discussion.html.

20. Coleman, SY, Gob, R, Manco, G, Pievatolo, A, Tort-Martorell, X, Reis, M (2016) How Can SMEs Benefit
from Big Data? Challenges and a Path Forward. Journal of Quality and Reliability Engineering Int.,
http://onlinelibrary.wiley.com/doi/10.1002/qre.2008/full.

21. Coleman, S.Y. Six Sigma – an opportunity for statistics and for statisticians. 2008, Vol. 5, pp. 94-96.

22. Mustafazade, F. (2018). Using social science data to solve a social housing problem.
https://blog.esrc.ac.uk/2018/10/19/using-social-science-data-to-solve-a-social-housing-problem/.

23. Smith, W, Coleman, S, Bacardit, J, Coxon, S. Insight from Data Analytics with an Automotive
Aftermarket SME. Quality and Reliability Engineering International. 2019, pp. 1-12.

24. Smith, W, Coleman, S, Bacardit, J. and Coxon, S. (2018) How data can change the automotive
aftermarket. Focus, p30-32, October, www.ciltuk.org.uk.

25. Zaman, I, Pazouki, K, Norman, R, Younessi, S, Coleman, SY. Development of automatic mode detection
system by implementing the statistical analysis of ship data to monitor the performance. Int. J. Maritime
Engineering, RINA Trans A3. 2017, Vol. 159, pp. 225-35.

26. Mustafazade, F, Coleman S, Bacardit, J (2018). Application of machine learning for decision support in
social housing. Statistics and Data Science - new Developments for Business and Industrial Applications
conference, Turin, http://www.sds2018.polito.it.

27. Provost, F., Fawcett, T. Data Science for Business - What you need to know about Data Mining and
Data-Analytic Thinking. s.l. : O’Reilly Media, USA., 2013.

28. Shan, Carl. What-are-good-examples-of-using-data-science-for-development-and-or-social-good.

quora.com. [Online] 2014. [Cited: 2 August 2019.] https://www.quora.com/What-are-good-examples-of-
using-data-science-for-development-and-or-social-good.

29. James, G, et al. The Introduction to Statistical Learning with Applications in R. s.l. : Springer, 2017.

30. Cady, Field. The Data Science Handbook, Wiley. s.l. : Wiley, 2017. ISBN: 978-1-119-09294-0.

17
 Vicario & Coleman, 2019

Grazia Vicario graduated in Mathematics at the University of Torino and is enrolled from 1998 as Associate
Professor of Statistics at the Department of Mathematical Sciences of Politecnico di Torino. Her main
research fields are reliability, design of experiment, statistical, quality control, design and analysis of
computer experiments. Her teaching experience includes courses of Statistics for Master and Bachelor
students, PhD Courses in Statistics for Experimenters and Master courses in Statistical Quality Control as
well as E-learning and Videoconference Courses in Basic, Applied and Advanced Statistics. Partnership in
European and Italian Research Framework Programs and Vice-President of European Network for Business
and Industrial Statistics (2009-2011 and 2015-2017).

Shirley Coleman studied mathematics and statistics before completing a PhD in Statistics and Computer
Graphics. She is Technical Director of the Industrial Statistics Research Unit at Newcastle University working
at the confluence of business, industry and academia in the application of statistics in data analytics,
monetising data and quality improvement. She is a Chartered Statistician and a Council member of the
Royal Statistical Society. She is an elected member of the International Statistics Institute and an active
member and former president of the European Network of Business and Industrial Statistics. Currently she
focusses on Innovate UK funded Knowledge Transfer Partnerships developing data science capabilities
within utilities and SMEs.

18