A Data-Science-as-a-Service Model

Matthias Pohl, Sascha Bosse and Klaus Turowski

Magdeburg Research and Competence Cluster,

Faculty of Computer Science, University of Magdeburg, Magdeburg, Germany.

Keywords:

Data-Science-as-a-Service, Cloud Computing, Service Model, Data Analytics, Data Science.

Abstract:

The keen interest in data analytics as well as the highly complex and time-consuming implementation lead to

an increasing demand for services of this kind. Several approaches claim to provide data analytics functions

as a service, however they do not process data analysis at all and provide only an infrastructure, a platform or

a software service. This paper presents a Data-Science-as-a-Service model that covers all of the related tasks

in data analytics and, in contrast to former technical considerations, takes a problem-centric and technology-

independent approach. The described model enables customers to categorize terms in data analytics environ-

ments.

1 INTRODUCTION

An increasing keen interest in data science and data

analytics exists as reported by a trend analysis of

the last 5 years (Google Trends). The data-intensive

world allures with generating revenue from analy-

zing information and data that are simply and quickly

available. Different approaches for knowledge dis-

covery (KDD) (Fayyad et al., 1996) or business-

related data mining (CRISP-DM) (Shearer, 2000)

are in use. However, proceeding is highly com-

plex, time-consuming and needs expertise in diffe-

rent disciplines, either computer sciences, mathema-

tics or a context-related specialization. The motiva-

tion within a company could arise from preventing

downtime of machines, getting insights about custo-

mer relationships or optimizing business processes.

If the required expertise for data analysis cannot be

provided internally, the tasks can be forwarded to

external consulting services. By using such servi-

ces it is possible to compensate the lack of exper-

tise, but it is very cost-intensive and still extremely

time-consuming. The paradigm of cloud computing

(Mell and Grance, 2011) establishes service concepts

that seem to be able to solve the remaining issues.

Among concepts like Infrastructure-as-a-Service

(IaaS), Platform-as-a-Service (PaaS) or Software-

as-a-Service (SaaS) that revolutionize computing

on several layers, service models like Analytics-

as-a-Service, Data-Analysis-as-a-Service, Business-

Analytics-as-a-Service or Big-Data-as-Service were

conceptualized. Whether these approaches are suit-

able recommendations, decision or application servi-

ces for non-expert customers or just analogies to stan-

dard concepts is not clariﬁed. Furthermore, a custo-

mer is confronted to choose within a set of apparently

unclear buzzwords like data science, data mining, big

data analytics, etc.

Therefore we make a predeﬁnition for the usage

of these terms and will diffuse a deﬁnition through

argumentation. The whole process that contains data

provision, data preparation, data analysis and data vi-

sualization is called data science in this paper. Data

analytics, business analytics and big data analytics are

often used synomously for data science, however they

could differ in context of use. Data mining is a key

term in most of related works and is used in varient

different ways. We will take this term as a byword for

data analysis.

This paper has twofold objectives. Firstly, it will

provide a full-service model that will be extracted

from existing approaches and will address data ana-

lytics services. Secondly, it will discuss the arising

service offers and data science process steps. The

subsumption of service models simpliﬁes the asses-

sment of IT service offers for companies that plan to

get insight from their data. From a scientiﬁc view,

a guideline is drawn for a future usage of terms in

data analytics environment and a classiﬁcation of past

work is a point of interest. The structure of the paper

is as follows. A knowledge base about related work

is presented in the second section. The review is built

up on a cross-reference search for data analytics ser-

vices and data mining process steps (Webster et al.,

432

Pohl, M., Bosse, S. and Turowski, K.

A Data-Science-as-a-Ser vice Model.

DOI: 10.5220/0006703104320439

In Proceedings of the 8th International Conference on Cloud Computing and Services Science (CLOSER 2018), pages 432-439

ISBN: 978-989-758-295-0

2002). The third section shows a Data-Science-as-

a-Service model that combines the different steps in

data mining as well as a cloud computing service mo-

del. Section four demonstrates examples of existing

services that are offered by known IT service provi-

ders. A discussion that argues aspects and challenges

from a technical and a scientiﬁc view as well as con-

cludes the paper.

2 RELATED WORKS

There are numerous works that address data mining,

data analytics, data science or similar. First of all, we

want to have a look on frameworks that are related

to data mining, data analytics and data science. In

case of cloud computing, different services that pro-

vide data analytics are already developed. Next to

data analytics services we will consider data science

process steps and focus on it in further search. In

order to retrieve relevant works, we took search en-

gine (Science Direct, Scopus) and knowledge databa-

ses (DBLP, IEEE, ACM) into account.

2.1 Data Science Frameworks

The data science process shall end up with know-

ledge. The key step in this nontrivial process is cal-

led data mining (Fayyad et al., 1996). Therefore data

has to be processed from selected sources and trans-

formed into a proper format. CRISP-DM (Shearer,

2000) describes a process cycle that is similar to Kno-

wledge Discovery in Databases (Fayyad et al., 1996),

but points out the business understanding that is ne-

cessary for initiating because data mining goals will

be deﬁned on that basis. Respecting both frameworks

it is conspicuous that data mining is a name of a pro-

cess step in KDD and entitles the whole reference

model CRISP-DM. In (Kurgan and Musilek, 2006),

a review of data mining frameworks is given. There

exist various derivates of the mentioned frameworks

with relation to a speciﬁc ﬁeld of application. In (Sun

et al., 2017), the authors add data analytics that also

contains data mining as a signiﬁcant part to the dis-

cussion and observe it as an aggregation of data ana-

lysis, data mining, data warehouse, statistic modeling

and data visualization. Furthermore the authors aim

at identifying an ontological separation between data

analytics, business analytics, knowledge analytics and

big data analytics. In (Nalchigar and Yu, 2017), a

conceptual modeling framework for business analy-

tics is proposed and provides catalogues for business

questions, analytic algorithms and data preparation.

In this manner, a transmission between business re-

quirements and algorithmic implementation is esta-

blished. With the evolution of a data-intensive world

new technologies are needed for handling rapidly gro-

wing, variously shaped mass of data (Laney, 2001).

In (Maltby, 2011), the author reviews big data with

relation to analytic techniques and mentions that data

mining combines statistics and machine learning with

database management. Machine learning is not a new

thing (Michalski et al., 1983) and focuses on ”au-

tomatically learn to recognize complex patterns and

make intelligent decisions based on data” (Maltby,

2011). Data analysis, machine learning and data vi-

sualization are marked as the core disciplines of data

science that is a new paradigm in the ﬁeld of big data

(Concolato and Chen, 2017). However, there is no

simple and uniﬁed data science framework (Marvasti

et al., 2015). Such frameworks are often deep into

speciﬁc context-related data and implemented in an

application environment (Brough et al., 2017; Loet-

sch and Ultsch, 2016). In (Cleveland, 2001), data ana-

lysis is described as an enlargement of statistics com-

bined with data computing and is called data science.

Obviously, there exists discordance about the terms

data science, (big) data analytics, data mining, etc.

However, it occurs that all deﬁnitions come up with

a similar structure. Data has to be selected, prepared

and analyzed to show up new information. However,

the needed expertise and the lack of uniformed pro-

cesses lead to the fact that data analytics is more an

art than science (Zorrilla and Garc

ıa-Saiz, 2013).

2.2 Data Science Services

The thought of automation without requiring human

interaction, the capability solving small and big data

problems, the availability as well as ﬂexibility in com-

puting power and storage characterize the essentials

of cloud computing (Mell and Grance, 2011). Dif-

ferent concepts that offering machines (IaaS), opera-

ting systems (PaaS) and applications (SaaS) are alre-

ady deﬁned and discussed (Dillon et al., 2010). There

exist approaches for hosting data analytics environ-

ments in a cloud. In (Xu et al., 2015), the authors aim

at making real-time data analytics available as a ser-

vice and deal with the challenges of creating service

interfaces, wrapping existing big data frameworks and

real-time processing of data. In (Zorrilla and Garc

ıa-

Saiz, 2013), an approach is motivated with some ex-

emplary services (e.g. GoodData, IBM Smart Analy-

tics) and it is concluded that none of them wraps the

whole analytic process that is described in KDD or

CRISP-DM. The proposed framework contains servi-

ces for the steps of data preparation, data analysis and

data visualization and is layer structured from an en-

A Data-Science-as-a-Service Model

433

terprise perspective. The concept is implemented as

a SaaS, though its automation is not sufﬁciently dis-

cussed. In (Ribeiro et al., 2015), the authors provide a

Machine-Learning-as-a-Service approach and discuss

the low ﬂexibility of implementing new algorithms on

existing platforms like PredictionIO or OpenCPU. In

(Xinhua et al., 2013), Big-Data-as-a-Service is deﬁ-

ned with getting insight from big data, which charac-

terizes this as big data analytics. In (Ardagna et al.,

2016), the authors describe the Big-Data-Analytics-

as-a-Service paradigm as suitable of companies do

not have enough data scientists. Along to service re-

quirements (data preparation, data analysis, data vi-

sualization), various challenges (data quality, diver-

sity, security, privacy) that come up with handling vast

data are examined by the authors. In (Grossman et al.,

2016) the variations of data science services (infra-

structure, platform, software, support) and a colloca-

tion of these is mentioned. The authors also describe

requirements like API-based access, data portability,

data peering and pay-for-compute, however automa-

tion in processing is not addressed. In (Medvedev

et al., 2017), an approach for setting up data mining

processes in a cloud environment is proposed and al-

lows a user to model a data mining process within the

steps of data uploading, data pre-processing, data mi-

ning and presentation. Several Analytics-as-a-Service

providers are compared in (Naous et al., 2017) with

respect to core features of data sources, data proces-

sing & preparation, analysis, visualizations and even

platform and infrastructure propositions. In conclu-

sion all provided services have a self-service charac-

ter.

2.3 Data Science Process Steps Services

After terms and existing data science services have

been discussed, we want to focus on the steps that

represent the parts of the data science process. Com-

panies are often interested in analyzing internal and

external business-related data. Within the scope of di-

gitalization, Internet-of-Things (IoT) applications are

in widespread use. In conjunction with that, a lot of

data services are formed that could have their origin

in data crawling concepts. In (Haase et al., 2011), an

information workbench for linked data applications is

provided, and a structured crawling system to gather

linked data over the web with standards like RDF is

proposed in (Isele et al., 2010). However such ser-

vices are known and described as Data-as-a-service

(Delen and Demirkan, 2013). In (Terzo et al., 2013),

Data-as-a-Service is interpreted as a storing and pro-

cessing service and the authors proposed a layer ar-

chitecture of an IaaS. In (Seibold and Kemper, 2012),

different types of Database-as-a-Service (shared ma-

chine, shared processes, shared tables) are described

and can appear as SaaS, PaaS and IaaS, which de-

pends on the complexity of the delivered systems. In

(Curino et al., 2011), the authors point out some pro-

perties (efﬁcient multi-tenancy, elastic scalability, and

database privacy) that have to be realized which is

conﬁrmed in (Agrawal et al., 2009). With respect to

data storing, data integration is similarly important.

In (Riedemann and Timm, 2003), the authors draw

the ”vision to achieve automated just-in-time integra-

tion”, and in (Bergamaschi et al., 2008) the idea co-

mes up that ”data integration systems is on producing

a comprehensive global schema successfully integra-

ting data from heterogeneous data sources”. In (Co-

hen and Richman, 2002), it is mentioned that data ma-

tching and clustering algorithms can create a solution

for these problems. At this point an intersection with

data preparation is obvious. ”Data preparation is an

important and critical step for complex data analysis”

(Yu et al., 2006). A handbook for data preparation is

provided in (Pyle, 1999) and describes the access of

data, data discovery and data modeling. It also focu-

ses on data mining and determines that ”the process

of data science is smooth and backward adjustment is

possible”. In (Yu et al., 2006), the authors mention

problems in data preparation like incomplete data,

noisy data, inconsistent data, selecting relevant data,

reducing data and resolving data conﬂicts, however,

they notices some solution approaches. In (Narman

et al., 2009), a model-based method for detecting

data accuracy problems is proposed. The straighte-

ned data has to be organized, so data modeling is

a further sub-process. In (Duggan and Yao, 2015),

the authors describe ”an approach [for] automating

most of this work, building data models from speci-

ﬁcations of a data collection system”.In (Song et al.,

2015), the authors address problems like combinato-

rial complexity, scattered modeling rules, semantic

mismatch, inexperience of novice designers, incom-

plete knowledge of designers and multiple solutions

in data modeling and give some solution techniques

that are categorized in linguistics based (e.g. NLP),

pattern-based, case-based, ontology-based and multi-

techniques-based. At this point, one can see a rela-

tion to machine learning techniques that can also be

used for outlier detection (Hawkins et al., 2002; Pru-

engkarn et al., 2016) or data matching (Rong et al.,

2012). The overall modeling concept could be na-

med as data warehousing. A transformation approach

from operational schemes to data warehouse is pre-

sented in (Dori et al., 2008). Such a data structure

changes with time and arising amount of data, so a

framework for real time data warehousing is suitable

CLOSER 2018 - 8th International Conference on Cloud Computing and Services Science

434

(Farooq and Sarwar, 2010). With the usage of ma-

chine learning techniques, one can see a connection to

data analysis tasks. The main aim is to automate the

task of selecting algorithms that analyze the data. In

(Garc

ıa-Saiz and Zorrilla, 2017), a framework is pre-

sented that includes a meta-learning approach. With

model-based, data contextual, information theoreti-

cal, complexity and statistical meta features a system

learns the way of appropriate algorithm selection for

a common problem. In (Luo, 2016), the author pro-

vides a literature review on automatic algorithm se-

lection for machine learning and categorizes different

approaches. It also exposes that the parameter and fe-

ature selection is an important part. In (Langley et al.,

1994), an approach with a heuristic search while in

(Hall, 2000) a correlation-based feature selection is

described. A different approach is introduced in (Es-

pinosa et al., 2013), in fact a taxonomy that provides a

recommendation about using data mining methods for

non-expert data miners. After these steps, a suitable

visualization or presentation of the results is neces-

sary. In (Matsushita et al., 2004), the authors focus on

an automated visualization of user required informa-

tion via processing data frames. In (Andrienko and

Andrienko, 1999), a data characterization scheme for

automated data visualization is provided. The cycle

between data analysis and data visualization is picked

up in (Wagner, 2015) where a process of reconﬁgu-

ring data analysis with ideas coming from visualiza-

tions is introduced. The essential challenges are con-

cluded in (Xu et al., 2015). The creation of service

interfaces that are suitable for integrating and proces-

sing distinct data sources in combination with analy-

zing frameworks and that are also utilizable for diffe-

rent applications is paired with the ability of real-time

responding.

3 DATA-SCIENCE-AS-A-SERVICE

MODEL

Considering prior works in the ﬁeld of data science

and correlated disciplines, a Data-Science-as-a-

Service model will be deduced in this section. The

majority of the existing approaches provide IaaS,

PaaS or SaaS models that enable users to conduct data

science. However, for serving data science there has

to be a service level above the known cloud compu-

ting models (Mell and Grance, 2011) that covers the

whole data science process. The earlier frameworks

and theoretical data mining approaches overlap in the

core steps of accessing data, arranging data, analyzing

data and presenting results. The automatization of the

process steps leads to separated services that can be

used as standalone ones. Therefore, we will describe

some entry and exit points of the sub-services to show

the feasibility of service separation. An entry point

is referred to a condition for initiating a sub-process.

An exit point is deﬁned as a point at which results are

available.

3.1 Data-as-a-Service

There have to exist data (Fayyad et al., 1996; Shearer,

2000) for a data science process. Either data is uplo-

aded or collected by a user (Naous et al., 2017; Med-

vedev et al., 2017) or integrated from data services

(Delen and Demirkan, 2013) in an adequate storage

system. The aim of this step called Data-as-a-Service

is providing a base of data that ideally cover all related

information. A ﬁrst entry point of this service model

is a chunk of data or an information request that could

be represented by keywords. On basis of the extracted

meta data or keywords related data could be provided

via data services. Hence, an exit point is the provision

of required data or an index-linked data pool that is,

for instance, stored in a traditional database or a dis-

tributed ﬁle system. It is shown that data services can

be connected even with other data sources and sup-

plied on appropriate infrastructure (Vu et al., 2012;

Isele et al., 2010; Bergamaschi et al., 2008).

3.2 Data-Preparation-as-a-Service

The process step that follows after data provision is

data preparation. It is termed as data selection, data

preprocessing, data transformation (Fayyad et al.,

1996), directly data preparation (Ardagna et al., 2016;

Naous et al., 2017; Shearer, 2000), data modeling or

generally data warehousing. However, all of these

terms are covered by Data-Preparation-as-a-Service.

Next to the previous service step a user-provided data

pool could also be an entry point for this service, ho-

wever, one has to consider that at least infrastructure

(IaaS) is needed. An exit point is a fully organized

set of data that is suitable for data analysis. In case of

unstructured or semi-structured data the data prepara-

tion service step is necessary before starting the data

analysis service. Following (Duggan and Yao, 2015;

Song et al., 2015; Yu et al., 2006; Narman et al., 2009)

it is possible to arrange data automatically after clean-

sing.

3.3 Data-Analysis-as-a-Service

The key step in the majority of the related approaches

is called data mining (Fayyad et al., 1996; Medvedev

et al., 2017; Shearer, 2000; Zorrilla and Garc

ıa-Saiz,

A Data-Science-as-a-Service Model

435

2013; Sun et al., 2017). Data analysis is also used for

indication (Xinhua et al., 2013; Naous et al., 2017;

Sun et al., 2017). The term data analytics that we in-

troduced as a synonym for data science is seldomly

mentioned (Ardagna et al., 2016), just like (machine)

learning and predicting (Ribeiro et al., 2015). Howe-

ver, all the approaches use the referred terms in sense

of Data-Analysis-as-a-Service. The concept of data

mining transposes the idea of creating data by pre-

dictive or prescriptive analysis. Nevertheless, in ge-

neral it is a derivation of data analysis. In a stand-

alone service consumption an entry point could be

a well-structured data set that will be used as input

data for analysis procedures and algorithms. An eva-

luated output (e.g. a data frame) is a possible exit

point. The automatic selection of algorithms (Luo,

2016) and a self-learning application system (Garc

ıa-

Saiz and Zorrilla, 2017) enable such a service.

3.4 Data-Visualization-as-a-Service

The last service step adresses data visualization which

is a term that has been chosen by most researchers

(Ardagna et al., 2016; Xinhua et al., 2013; Naous

et al., 2017; Sun et al., 2017; Zorrilla and Garc

ıa-Saiz,

2013). However, an interpretation of results (Fayyad

et al., 1996), a deployment (Shearer, 2000) or a struc-

tured output for a customers product (Xu et al., 2015)

is mentioned. An entry point could be structured data

that should only be visualized. However, a service for

distributing analysis results is also conceivable that

could be seen as the exit point. Approaches by (Mat-

sushita et al., 2004) and (Andrienko and Andrienko,

1999) can facilitate a visualization service.

Data-Analysis-as-a-Service

Data-Visualizaon-as-a-Service

Data-as-a-Service

Data-Preparaon-as-a-Service

Soware-as-a-Service

Plaorm-as-a-Service

Infrastructure-as-a-Service

Figure 1: Data-Science-as-a-Service model.

3.5 Data-Science-as-a-Service

The Data-Science-as-a-Service model (Fig. 1) com-

bines the explained services and realizes the whole

data science process. Data-Science-as-a-Service is

the ordered sequence of Data-as-a-Service, Data-

Preparation-as-a-Service, Data-Analysis-as-a-Service

and Data-Visualization-as-a-Service. The entry

points of Data-Science-as-a-Service are equal to the

ones of Data-as-a-Service. Exit points could be de-

rived from Data-Visualization-as-Service. Conside-

ring the model a combination that does not contain all

of the sub-processes is also conceivable. The entry

and exit points of the sub-services enable the usage

of sub-sequences. For instance, one could getting

started with a information requirement at Data-as-

a-Service and quit with a well-structured dataset at

Data-Preparation-as-a-Service.

Furthermore, the idea of backward adjustment

(Pyle, 1999; Wagner, 2015) is included. In case a

service result would not fulﬁll the customer require-

ments a service step could be re-processed or a previ-

ous service step could be invoked. For instance, if the

output result of the data analysis service is not suita-

ble or applicable then the step could be repeated. At

this point the service system gets the possibility to le-

arn and can forward the demand of advancement if it

is not successful. Otherwise the system rolls back to

the underlying Data-Preparation-as-a-Service to rear-

range the data (e.g. new metrics) to gain better results

in the next step. This functionality is only useable in

case of requesting more than one service.

The essential characteristics of cloud computing

(Mell and Grance, 2011) are necessary for provi-

ding Data-Science-as-a-Service. Although the data-

base services or data integration services are part of

Data-as-a-Service these services have to be supported

by infrastructure, platform and software to process all

of the service steps. At this point we do not want to

categorize the underlying services. Nevertheless, re-

source elasticity, service measurement and broad net-

work access is required to offer Data-Science-as-a-

Service.

Data analytics could be used as a similar expres-

sion for data science. Normally the initiation of a

data science process is forced by a user or business

requirements that expect some kind of knowledge or

insight from data. Thus, the terms business analy-

tics, Insight-as-a-Service or Knowledge-as-a-Service

(Terzo et al., 2013) are used. In case of big data pro-

blems, there exist different approaches about so cal-

led Big-Data-Analytics-as-a-Service or Big-Data-as-

a-Service. However, all of these concepts are covered

by the presented Data-Science-as-a-Service concept.

CLOSER 2018 - 8th International Conference on Cloud Computing and Services Science

436

4 EXAMPLE

In the previous section we presented a Data-Science-

as-a-Service model. In (Naous et al., 2017), it is

shown that all of the data science cloud computing

offers (e.g. Google Cloud Platform, IBM Bluemix,

SAP Cloud Platform, Amazon Quicksight or Good-

Data Platform) are some kind of self-service via soft-

ware or platform. The GoodData and SAP Cloud

platform cover the most service steps, from data pro-

vision up to visualization. However, if one has a

look on the current product line of Google Services

all data science process steps are covered (e.g. Big-

Query, CloudStorage, DataFlow, TensorFlow, Pre-

diction API, Charts and Firebase). Nevertheless none

of the listed providers offers a completely data science

service that proceeds (nearly) automatically.

Observing Google’s image recognition tools we

ﬁnd a demonstrating example for Data-Science-as-

a-Service. For instance, Google Goggles combines

the whole sequence of the Data-Science-as-a-Service

model. The starting point of the service is a taken

picture of a user. Google provide databases with

website links and images as Data-as-a-Service. The

image of a user has to be transformed in a suita-

ble format for further processing (Data-Preparation-

as-a-Service). Google’s machine learning algorithms

will detect patterns of interest by means of, for in-

stance, trained neural networks (Data-Analysis-as-

Servce). Afterwards a presentation of the detected

image areas and referred website links is given (Data-

Presentation-as-a-Service). Furthermore, the service

can learn from a rating of the result regarding if a user

is satisﬁed or not. In the unsatisﬁed case enriching

the databases or implementing new algorithms could

be ﬁtting solutions. New computation methods may

assume reprepared input data. Although, new data

deﬁnitely involve a repreparation. Hence, we observe

the formerly mentioned backward adjustment.

5 DISCUSSION AND

CONCLUSION

There is a many-faceted choice of products that range

from IaaS over PaaS to SaaS. Providing a platform or

a software that allows to conduct data science is not

Data-Science-as-a-Service, however such an offer is

not possible without. Data-Science-as-a-Service is a

symbiosis of infrastructure, platform, software and

the processing of data science tasks. This can be done

automatically or semi-automatically, e.g. with opti-

ons of user interactions for launching further service

steps or re-processing. Otherwise there would

not be an added value for a customer in compari-

son to common services. Considering (Vargo and

Lusch, 2004), value only results from the beneﬁcial

application of data science services where the infra-

structure is a transmitter. Furthermore, the essential

characteristics of cloud computing like on-demand

self-service, broad network access, resource pooling,

rapid elasticity and measured service are given. The

same refers to the deployment models. Considering

the service measurement one could think about new

business models that specify a typical pay-per-use

model. Among a pay-per-compute, a pay-per-insight

model is potentially conceivable. Measurements

could also differ on service levels. Ultimately, it is

an existing future task, because it drops the question

how to measure insight in that case. We have shown

that a data science service system is feasible with

the connection of several service steps. However

we describe an overall service model and not the

orchestration of services in detail. The connection

between the service steps has to be observed in

future. There exist different frameworks (e.g. RDF)

or markup languages (e.g. PMML) that can be used

to determine utility. Challenges will arise to achieve

smooth transitions, even though when a service

step is skipped by a customer or the functionality

of backward adjustment is applied. Therefore it is

not even called service layers, because the usage as

stand-alone services is also possible. ”Information

systems that provide such capabilities are often called

business intelligence tools nowadays” (Delen and

Demirkan, 2013). Indeed one can see parallels to

business intelligence (BI). Comparable concepts are

given in the theory of decision support systems (DSS)

that could be included in future research. Especially

BI tools focus on a technical layer structure for col-

lecting business data, re-arrangement and reporting.

However, there also exists an alternate deﬁnition

that sees analytics as a subset of BI (Davenport

et al., 2001). From a point of service a data/business

understanding is not necessary. The customer is

forced to input its information demand and to check

the knowledge outcome on its suggestions. If an

automated data science service system is able to

govern requirements and in-depth knowledge one

might call it Artiﬁcial Intelligence. Associating

decision support decision model presentation could

be included in Data-Presentation-as-a-Service in

future. With the given service models that orientate

towards the key steps of data mining it is possible to

characterize terms that occurs in the analytics service

ﬁeld. One has only to decide if a term is related to

a sub-process or the whole system. Provisioning

A Data-Science-as-a-Service Model

437

an ontology in the ﬁeld of data analytics will be a

prospective aim.

In this paper, a service model framework for data

science was created. It enables business and scientiﬁc

customers to classify offers of common data science

services and to substantiate their expectations. The

results are furthermore useful as a template for crea-

ting data science services.

REFERENCES

Agrawal, D., El Abbadi, A., Emekci, F., and Metwally, A.

(2009). Database management as a service: Challen-

ges and opportunities. In Data Engineering, 2009.

ICDE’09. IEEE 25th International Conference on, pa-

ges 1709–1716. IEEE.

Andrienko, G. and Andrienko, N. (1999). Data characte-

rization schema for intelligent support in visual data

analysis. Lecture notes in computer science, pages

349–366.

Ardagna, C. A., Ceravolo, P., and Damiani, E. (2016). Big

data analytics as-a-service: Issues and challenges. In

Big Data (Big Data), 2016 IEEE International Confe-

rence on, pages 3638–3644. IEEE.

Bergamaschi, S., Po, L., and Sorrentino, S. (2008). Auto-

matic annotation for mapping discovery in data inte-

gration systems. In SEBD, volume 2008, pages 334–

341.

Brough, D. B., Wheeler, D., and Kalidindi, S. R. (2017).

Materials knowledge systems in python—a data

science framework for accelerated development of

hierarchical materials. Integrating Materials and Ma-

nufacturing Innovation, 6(1):36–53.

Cleveland, W. S. (2001). Data science: an action plan for

expanding the technical areas of the ﬁeld of statistics.

International statistical review, 69(1):21–26.

Cohen, W. W. and Richman, J. (2002). Learning to ma-

tch and cluster large high-dimensional data sets for

data integration. In Proceedings of the eighth ACM

SIGKDD international conference on Knowledge dis-

covery and data mining, pages 475–480. ACM.

Concolato, C. E. and Chen, L. M. (2017). Data science: A

new paradigm in the age of big-data science and ana-

lytics. New Mathematics and Natural Computation,

13(02):119–143.

Curino, C., Jones, E. P., Popa, R. A., Malviya, N., Wu,

E., Madden, S., Balakrishnan, H., and Zeldovich, N.

(2011). Relational cloud: A database-as-a-service for

the cloud. In 5th Biennial Conference on Innovative

Data Systems Research, pages 235–240. MIT.

Davenport, T. H., Harris, J. G., David, W., and Jacobson,

A. L. (2001). Data to knowledge to results: buil-

ding an analytic capability. California Management

Review, 43(2):117–138.

Delen, D. and Demirkan, H. (2013). Data, information

and analytics as services. Decision Support Systems,

1(55):359–363.

Dillon, T., Wu, C., and Chang, E. (2010). Cloud compu-

ting: issues and challenges. In Advanced Information

Networking and Applications (AINA), 2010 24th IEEE

International Conference on, pages 27–33. IEEE.

Dori, D., Feldman, R., and Sturm, A. (2008). From con-

ceptual models to schemata: An object-process-based

data warehouse construction method. Information Sy-

stems, 33(6):567–593.

Duggan, D. and Yao, J. (2015). Automated data modeling

for health data collection. Stevens Institute Technical

Report.

Espinosa, R., Garc

ıa-Saiz, D., Zorrilla, M., Zubcoff, J. J.,

and Maz

on, J.-N. (2013). Enabling non-expert users

to apply data mining for bridging the big data divide.

In International Symposium on Data-Driven Process

Discovery and Analysis, pages 65–86. Springer.

Farooq, F. and Sarwar, S. M. (2010). Real-time data ware-

housing for business intelligence. In Proceedings of

the 8th International Conference on Frontiers of In-

formation Technology, page 38. ACM.

Fayyad, U. M., Piatetsky-Shapiro, G., Smyth, P., et al.

(1996). Knowledge discovery and data mining: To-

wards a unifying framework. In KDD, volume 96, pa-

ges 82–88.

Garc

ıa-Saiz, D. and Zorrilla, M. (2017). A meta-learning

based framework for building algorithm recommen-

ders: An application for educational arena. Journal of

Intelligent & Fuzzy Systems, 32(2):1449–1459.

Grossman, R. L., Heath, A., Murphy, M., Patterson, M., and

Wells, W. (2016). A case for data commons: Toward

data science as a service. Computing in Science &

Engineering, 18(5):10–20.

Haase, P., Schmidt, M., and Schwarte, A. (2011). The infor-

mation workbench as a self-service platform for lin-

ked data applications. In Proceedings of the Second

International Conference on Consuming Linked Data-

Volume 782, pages 119–124. CEUR-WS.org.

Hall, M. A. (2000). Correlation-based feature selection for

discrete and numeric class machine learning. In Pro-

ceedings of the Seventeenth International Conference

on Machine Learning, pages 359–366. Morgan Kauf-

mann Publishers Inc.

Hawkins, S., He, H., Williams, G., and Baxter, R. (2002).

Outlier detection using replicator neural networks. In

DaWaK, volume 2454, pages 170–180. Springer.

Isele, R., Umbrich, J., Bizer, C., and Harth, A. (2010). Ld-

spider: An open-source crawling framework for the

web of linked data. In Proceedings of the 2010 In-

ternational Conference on Posters & Demonstrations

Track-Volume 658, pages 29–32. CEUR-WS.org.

Kurgan, L. A. and Musilek, P. (2006). A survey of know-

ledge discovery and data mining process models. The

Knowledge Engineering Review, 21(1):124.

Laney, D. (2001). 3d data management: Controlling data

volume, velocity and variety. META Group Research

Note, 6:70.

Langley, P. et al. (1994). Selection of relevant features in

machine learning. In Proceedings of the AAAI Fall

symposium on relevance, volume 184, pages 245–271.

CLOSER 2018 - 8th International Conference on Cloud Computing and Services Science

438

Loetsch, J. and Ultsch, A. (2016). Process pharmacology:

A pharmacological data science approach to drug de-

velopment and therapy. CPT: Pharmacometrics & Sy-

stems Pharmacology, 5(4):192–200.

Luo, G. (2016). A review of automatic selection methods

for machine learning algorithms and hyper-parameter

values. Network Modeling Analysis in Health Infor-

matics and Bioinformatics, 5(1):18.

Maltby, D. (2011). Big data analytics. In 74th Annual Meet-

ing of the Association for Information Science and

Technology (ASIST), pages 1–6.

Marvasti, M. A., Poghosyan, A. V., Harutyunyan, A. N.,

and Grigoryan, N. M. (2015). Ranking and updating

beliefs based on user feedback: Industrial use cases.

In 2015 IEEE International Conference on Autonomic

Computing, pages 227–230.

Matsushita, M., Maeda, E., and Kato, T. (2004). An inte-

ractive visualization method of numerical data based

on natural language requirements. International jour-

nal of human-computer studies, 60(4):469–488.

Medvedev, V., Kurasova, O., Bernatavi

cien

e, J., Treigys, P.,

Marcinkevi

cius, V., and Dzemyda, G. (2017). A new

web-based solution for modelling data mining proces-

ses. Simulation Modelling Practice and Theory.

Mell, P. and Grance, T. (2011). The NIST deﬁnition of

cloud computing. NIST Special Publication, 800:145.

Michalski, S. R., Carbonell, G. J., and Mitchell, M. T.

(1983). Machine learning: An artiﬁcial intelligence

approach. Understanding the Nature of Learning,

2:3–26.

Nalchigar, S. and Yu, E. (2017). Conceptual modeling for

business analytics: A framework and potential bene-

ﬁts. In Business Informatics (CBI), 2017 IEEE 19th

Conference on, volume 1, pages 369–378. IEEE.

Naous, D., Schwarz, J., and Legner, C. (2017). Analytics as

a service: Cloud computing and the transformation of

business analytics business models and ecosystems. In

Proceedings of the 25th European Conference on In-

formation Systems (ECIS), Guimares, Portugal, pages

487–501.

Narman, P., Johnson, P., Ekstedt, M., Chenine, M., and Ko-

nig, J. (2009). Enterprise architecture analysis for data

accuracy assessments. In Enterprise Distributed Ob-

ject Computing Conference, 2009. EDOC’09. IEEE

International, pages 24–33. IEEE.

Pruengkarn, R., Wong, K. W., and Fung, C. C. (2016).

Data cleaning using complementary fuzzy support

vector machine technique. In International Confe-

rence on Neural Information Processing, pages 160–

167. Springer.

Pyle, D. (1999). Data preparation for data mining, vo-

lume 1. Morgan Kaufmann.

Ribeiro, M., Grolinger, K., and Capretz, M. A. (2015).

Mlaas: Machine learning as a service. In Machine

Learning and Applications (ICMLA), 2015 IEEE 14th

International Conference on, pages 896–902. IEEE.

Riedemann, C. and Timm, C. (2003). Services for data in-

tegration. Data Science Journal, 2:90–99.

Rong, S., Niu, X., Xiang, E. W., Wang, H., Yang, Q., and

Yu, Y. (2012). A machine learning approach for in-

stance matching based on similarity metrics. In Inter-

national Semantic Web Conference, pages 460–475.

Springer.

Seibold, M. and Kemper, A. (2012). Database as a service.

Datenbank-Spektrum, 12(1):59–62.

Shearer, C. (2000). The crisp-dm model: the new blue-

print for data mining. Journal of data warehousing,

5(4):13–22.

Song, I.-Y., Zhu, Y., Ceong, H., and Thonggoom, O. (2015).

Methodologies for semi-automated conceptual data

modeling from requirements. In International Con-

ference on Conceptual Modeling, pages 18–31. Sprin-

ger.

Sun, Z., Strang, K., and Firmin, S. (2017). Business

analytics-based enterprise information systems. Jour-

nal of Computer Information Systems, 57(2):169–178.

Terzo, O., Ruiu, P., Bucci, E., and Xhafa, F. (2013). Data

as a service (daas) for sharing and processing of large

data collections in the cloud. In Complex, Intelligent,

and Software Intensive Systems (CISIS), 2013 Seventh

International Conference on, pages 475–480. IEEE.

Vargo, S. L. and Lusch, R. F. (2004). Evolving to a new

dominant logic for marketing. Journal of marketing,

68(1):1–17.

Vu, Q. H., Pham, T.-V., Truong, H.-L., Dustdar, S., and

Asal, R. (2012). Demods: A description model for

data-as-a-service. In Advanced Information Networ-

king and Applications (AINA), 2012 IEEE 26th Inter-

national Conference on, pages 605–612. IEEE.

Wagner, M. (2015). Integrating explicit knowledge in the

visual analytics process. Doctoral Consortium on

Computer Vision, Imaging and Computer Graphics

Theory and Applications (DCVISIGRAPP 2015), Sci-

tepress Digital Library.

Webster, J., Kingston, O., and Watson, R. T. (2002). Ana-

lyzing the past to prepare for the future: Writing a

literature review. MIS Quarterly, 26(2):xiii–xxiii.

Xinhua, E., Han, J., Wang, Y., and Liu, L. (2013). Big data-

as-a-service: Deﬁnition and architecture. In Commu-

nication Technology (ICCT), 2013 15th IEEE Interna-

tional Conference on, pages 738–742. IEEE.

Xu, D., Wu, D., Xu, X., Zhu, L., and Bass, L. (2015). Ma-

king real time data analytics available as a service. In

Proceedings of the 11th International ACM SIGSOFT

Conference on Quality of Software Architectures, pa-

ges 73–82. ACM.

Yu, L., Wang, S., and Lai, K. K. (2006). An integrated data

preparation scheme for neural network data analysis.

IEEE Transactions on Knowledge and Data Engineer-

ing, 18(2):217–230.

Zorrilla, M. and Garc

ıa-Saiz, D. (2013). A service orien-

ted architecture to provide data mining services for

non-expert data miners. Decision Support Systems,

55(1):399–411.

A Data-Science-as-a-Service Model

439