Development of a Context-Free Data Ingestion Mechanism for AutoML

Gabriel Mac’Hamilton

and Alexandre M. A. Maciel

Universidade de Pernambuco, Recife, Brazil

Keywords:

Data Ingestion, AutoML, Machine Learning, Human-Computer Interaction.

Abstract:

Automated Machine Learning (AutoML) is a technology that simpliﬁes complex data processing and analysis

for strategic decision-making by automating machine learning tasks and enhancing the user experience. Data

ingestion is a crucial AutoML step that involves collecting external data for machine learning workﬂows. Typ-

ically, AutoML systems include data input modules. However, the lack of a user interface limits the number

of users that can utilize it. This work presents the development of a data ingestion mechanism that stream-

lines and simpliﬁes this machine learning stage into an AutoML framework called FMD. The mechanism

underwent three validations: Experimentation in a real-world scenario with two databases from different con-

texts, evaluation from expert opinions, and usability assessment through a questionnaire using the AttrakDiff

method. Following the validations, successful results were achieved in both assessments and in demonstrating

the ingestion in various contexts.

1 INTRODUCTION

Amidst an increasingly complex environment, where

the shortage of data scientists becomes more evident

in the face of high market and academy demands, Au-

toML systems emerge as a strategic solution for or-

ganizations. These systems can reduce the complex-

ity of the model building and optimization processes

(Hutter et al., 2019). In this context, AutoML is not

just a response to human resource limitations but also

a strategy to optimize the potential of available data,

accelerate model development cycles, and democra-

tize data science within organizations while also al-

lowing domain experts to work directly with machine

learning development (Elshawi et al., 2019).

Data ingestion is of crucial relevance in data sci-

ence, serving the purpose of consistently and reliably

introducing data into the machine learning pipeline

(Hapke and Nelson, 2020). According to studies, data

wrangling activities, encompassing ingestion, clean-

ing, and data transformation, consume about 70%

of data scientists’ time (Saurav and Schwarz, 2016).

This underscores the signiﬁcance of simplifying data

ingestion for AutoML systems, as it helps reduce time

and effort in this workﬂow phase. Such simpliﬁcation

allows users to dedicate more time to metadata prepa-

ration and model reﬁnement activities, contributing to

https://orcid.org/0000-0002-3735-190X

https://orcid.org/0000-0003-4348-9291

increased efﬁciency in machine learning teams (Patel,

2020).

Given the context, this work presents a data in-

gestion mechanism to alleviate the problems related

to the early stages of machine learning practices, do-

main understanding, and feature engineering. The

mechanism proposes a standardized and simpliﬁed

way to ingest data, using a metadata ﬁle called “con-

text ﬁle”, to help with transforming domain experts’

tacit knowledge into explicit knowledge, allowing for

better feature selection, allied with a robust data in-

gestion engine based on the ETL (extract, transform

& load) process, and a simple and intuitive user inter-

face (UI).

2 BACKGROUND

2.1 AutoML

AutoML is broadly deﬁned in the literature and

commonly describes systems that automate machine

learning activities. These systems were motivated

by the need to address the limitations and challenges

posed by traditional machine learning techniques.

These challenges include requiring highly specialized

professionals in model development, dependence on

domain experts’ knowledge tied to a particular model,

and introducing human biases, making the models in-

Mac’Hamilton, G. and Maciel, A. M. A.

Development of a Context-Free Data Ingestion Mechanism for AutoML.

DOI: 10.5220/0013357700003929

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 27th International Conference on Enterprise Information Systems (ICEIS 2025) - Volume 2, pages 581-588

ISBN: 978-989-758-749-8; ISSN: 2184-4992

581

efﬁcient. AutoML systems can expedite the develop-

ment of machine learning models, ensure greater opti-

mization, and democratize access to data science for a

broader and less specialized audience (Nagarajah and

Poravi, 2019). In summary, AutoML aims to ﬁnd an

optimized solution for machine learning applications

(Chen et al., 2021).

AutoML systems can automate any stage of ma-

chine learning model development. However, accord-

ing to (Hutter et al., 2019), most systems focus on

preprocessing and model tuning, emphasizing hyper-

parameter optimization, meta-learning, and neural ar-

chitecture search. The authors also highlight several

advantages of using a system that automates these ac-

tivities, such as reducing human effort in model devel-

opment, improving algorithm performance, enhanc-

ing reproducibility in academic work, and facilitating

the reuse of successful models.

2.2 Data Ingestion

The data ingestion is primarily characterized by ob-

taining and transporting data from an external source

to the machine learning workﬂow. Organizations typ-

ically employ this process to optimize data collection,

improve data quality and accuracy, and save time and

resources. Through the use of data ingestion tech-

niques, it is possible to reduce costly errors in the data

collection stage (Hapke and Nelson, 2020).

The main goal of this process is to capture, store,

and make data available for future use. Among the

methods found in the literature for developing data in-

gestion, we can highlight batch and streaming. Batch

data ingestion is usually performed through ETL (Ex-

tract, Transform & Load) routines that collect data

from an external source, incorporate it into the work-

ﬂow, or store it for later use. The batch technique

is employed for data that does not need to be con-

sumed in real time. Streaming data ingestion, on the

other hand, is used in cases where there is a need for

real-time data consumption, requiring speciﬁc tech-

nologies to support this type of demand (Hlupic and

Punis, 2021).

2.3 Software Usability

The graphic interface design of software can deter-

mine the success or failure of a product. A tool needs

a user-friendly interface to gain user approval and

may be replaced by competing options. Therefore,

the application of efﬁcient user experience (UX) and

user interface (UI) techniques is highly relevant, al-

lowing the development of a valuable system for users

(Tidwell et al., 2020).

The efﬁciency of a UI depends on its intuitive-

ness and ease of use. Intuitive software is designed to

be familiar, with recognizable components and pre-

cise interactions, enabling users to apply their prior

knowledge and use the interface seamlessly. Due

to the need for this familiarity, design patterns are

encouraged, allowing users to easily recognize the

functionalities displayed in interfaces (Tidwell et al.,

2020). In addition to interface familiarity, the over-

all user experience is a relevant factor in software

development. UX is about the visual interface and

how users perceive, interpret, and interact with a sys-

tem. Whalen (2019) emphasizes the need to consider

cognitive psychology, thinking patterns, and user ex-

pectations when creating effective designs (Whalen,

2019).

2.4 Data Mining Framework - FMD

The Data Mining Framework - FMD is an AutoML

system originally developed in 2017 at the University

of Pernambuco through various academic works. This

achievement is due to its nature as an open-source

software, which has been enhanced through multiple

projects over time.

The project was initially conceived to enable data

mining in virtual learning environments (VLE), with

the goal of democratizing data mining activities for

users with limited technical knowledge in this ﬁeld.

It was initially named as Visual Educational Data

Mining Framework - FMDEV, later being changed to

simply Data Mining Framework - FMD. The frame-

work allowed data mining from the Moodle VLE

with just a few clicks, providing data analytics and

visual graphs to users. The project was developed

using technologies such as Hypertext Markup Lan-

guage (HTML), Cascading Style Sheets (CSS), and

JavaScript, and was integrated into Moodle as an

HTML block (Gonc¸alves et al., 2017). The initial ar-

chitecture of FMD is represented in Figure 1.

The work of (da Silva, 2020) enhanced the project,

leading to its current state with a more robust architec-

ture, updated technologies, new functionalities, and

a user-friendly interface. The project underwent a

refactoring process using Lean Inception techniques

and requirements engineering, along with a technol-

ogy survey through an systematic literature review

(SLR). This allowed the development of a more re-

ﬁned architecture based on the original design. At this

moment, the project transitioned from being a data

mining framework to becoming an AutoML frame-

work, capable of performing automated supervised

https://moodle.org/

ICEIS 2025 - 27th International Conference on Enterprise Information Systems

582

machine learning tasks and presenting the results vi-

sually, using data related to the educational context.

Figure 1: FMD original architecture.

2.5 Related Work

This section presents a comparative analysis of open-

source AutoML tools in the market, focusing on us-

ability and data ingestion. Speciﬁcally, it examines

functionalities identiﬁed as essential by (Alves and

Maciel, 2023). To select the AutoML tools for com-

parison, those in the work of (Z

oller and Huber,

2021) were considered based on their number of ci-

tations and stars on GitHub. The ﬁve frameworks se-

lected for benchmarking were: TPOT

, hpsklearn

auto-sklearn

, H2O AutoML

, and ATM

(Auto Tune

Models).

Table 1 compares this project regarding data in-

gestion with the ﬁve most relevant open-source Au-

toML tools. Observing the table, it becomes evident

that, due to their nature being run from the command

line, TPOT, auto-sklearn, hpsklearn, and ATM lack

most of the functionalities described as necessary for

easy data ingestion. Being the only one with a graph-

ical interface, H2O AutoML has the most functionali-

ties. However, the interface primarily focuses on Au-

toML rather than data ingestion, and data input is lim-

ited to data ﬁles only. The FMD Data Ingestor stands

out as it was developed with a speciﬁc focus on the

most relevant functionalities outlined in the literature

for data ingestion, allowing data input through ﬁles or

database connections but lacking metadata inference.

This absence is mitigated by the “context mapping”

functionality, which pre-provides the necessary meta-

https://epistasislab.github.io/tpot/

https://hyperopt.github.io/hyperopt-sklearn/

https://github.com/automl/auto-sklearn

https://h2o.ai/

https://hdi-project.github.io/ATM/

data for speciﬁc datasets.

Table 1: AutoML Frameworks Benchmark.

Framework User Data Multiple data Metadata

InterfaceVisualization inputs Inference

FMD Yes Yes Yes No

TPOT No No Yes No

hpsklearn No No Yes No

auto-sklearn No No Yes No

H2O AutoML Yes Yes No Yes

ATM No No Yes No

3 METHODOLOGY

The project’s requirements gathering was conducted

using Design Science Research (DSR) (Aken, 2004)

techniques combined with lean inception (Caroli,

2018) and traditional methods of software require-

ments documentation. Several meetings were held

with project stakeholders for brainstorming and arti-

fact validation, such as user journeys, mockups, and

prototypes. Two personas were considered for deﬁn-

ing functionalities: the domain expert and the data

scientist.

Given that the developed project is premised on

integration with an existing AutoML solution, the

FMD, the new project architecture must complement

the one previously used. Thus, a new layer of data

ingestion and processing was added to the original ar-

chitecture deﬁned by (da Silva, 2020). The project

utilizes Javascript for the frontend layer and Python

for the backend layer, with an additional data inges-

tion layer managed by the Pentaho Data Integration

Community Edition

(PDI-CE) platform, invoked by

the backend. A visual representation of the project

architecture is presented in Figure 2.

The developed platform presents two distinct and

complementary workﬂows: the “context ﬁle” regis-

tration and the data source registration as presented

on Figure 3. A “context ﬁle” is represented by a

JSON-format ﬁle and contains the necessary meta-

data for analyzing data from a speciﬁc domain. There

are two options for data source registration: one for

registering data sources from CSV ﬁles and another

for connecting to a PostgreSQL, MySQL, or Oracle

database. After completing the data source registra-

tion, the data ingestion begins automatically.

https://www.hitachivantara.com/en-

us/products/pentaho-plus-platform/data-integration-

analytics/pentaho-community-edition.html

Development of a Context-Free Data Ingestion Mechanism for AutoML

583

Figure 2: Project architecture.

Figure 3: Project workﬂow.

4 RESULTS

4.1 Data Ingestion Interface for

AutoML

Following the developed and validated prototypes, a

graphical user interface for data ingestion was cre-

ated. According to the project workﬂow, described

in Section 3, screens were developed for uploading

“Context Files” and for conﬁguring data ingestion.

Figure 4 presents a screen for registering and up-

loading the context ﬁle by a Domain Expert user.

After the upload, the transformation json storage.ktr

(Figure 5) is executed, which loads the “Context File”

into the project’s Data Lake, making it available for

selection in the data ingestion conﬁguration area. Be-

sides the upload, the user can also edit the context on

the screen.

Figure 4: User Interface for context ﬁle upload.

Figure 5: Transformation json storage.ktr viewed at the vi-

sual interface of the PDI-CE.

Figure 6 represents the main screen for registering

datasets, where the user will conﬁgure data ingestion

in two steps, in a wizard interface format that repre-

sents a step-by-step guide for easy conﬁguration.

Figure 6: User Interface for the Data Ingestion Conﬁgura-

tion.

After the conﬁguration, a series of transforma-

ICEIS 2025 - 27th International Conference on Enterprise Information Systems

584

tions are executed in the backend, which collect the

registered information, gather the data, and execute

the data ingestion into the project’s Data Lake. By

default, the data is stored in CSV format for bet-

ter interaction with the AutoML platform; therefore,

data transformations are required if it originates from

a database connection, performed by the PDI. The

ﬁrst transformation executed is cria headers csv.ktr

(Figure 7) that generates the CSV metadata for the

data ﬁle produced at the end of the process. Next,

the database data collection transformation, called in-

gestor bd.ktr (Figure 8), is executed, and ﬁnally, the

data is loaded into the Data Lake in CSV format by

the transformation ingestor bd carga.ktr (Figure 9).

Figure 7: Transformation cria headers csv.ktr viewed at the

visual interface of the PDI-CE.

Figure 8: Transformation ingestor bd.ktr viewed at the vi-

sual interface of the PDI-CE.

Figure 9: Transformation ingestor bd carga.ktr viewed at

the visual interface of the PDI-CE.

4.2 Data Ingestion in a Real-World

Scenario

Three context ﬁles were created to demonstrate data

ingestion capabilities with different data contexts.

These contexts are based on existing work in the lit-

erature that shows the best features for analyzing spe-

ciﬁc themes. The themes were: breast tumor classiﬁ-

cation using the six most important attributes, breast

tumor classiﬁcation using the ten most important at-

tributes, and analysis of student engagement in virtual

learning environments.

The two breast cancer contexts were used to

demonstrate the ability to use various contexts with

the same database; they were also added to the plat-

form through a database connection. The engagement

dataset, which was ingested into the platform from

a CSV ﬁle, demonstrated the platform’s capacity to

work with widely different contexts (education and

health).

The deﬁnition of the two contexts related to breast

cancer was based on the work of (Ray et al., 2020),

which aimed to analyze the best characteristics of the

analysis of breast cancer cells to determine whether

they are malignant or benign. The same dataset used

in that work, the Wisconsin Diagnosis Breast Can-

cer dataset (WDBC), containing 573 instances, was

used for data ingestion. The deﬁnition of the engage-

ment context was inspired by (H. R. Mac

edo et al.,

2021), which sought to determine the most relevant

data for analyzing student engagement proﬁles in vir-

tual learning environments. The engagement dataset

was provided by Research Group in Data Science

and Analytics (GPCDA), which comprised 30,217 in-

stances. After data ingestion and AutoML execution,

the results are shown in Table 2, being (I) Top 6 fea-

tures for Breast Cancer context, (II) Top 10 features

for Breast Cancer context and (III) Features for stu-

dents engagement context.

Table 2: Performance Metrics Results.

Context Accuracy AUC Recall Precision F1 Score

(I) 0.91 0.96 0.93 0.93 0.93

(II) 0.93 0.99 0.91 0.99 0.95

(III) 0.94 0.99 0.92 0.94 0.94

The AutoML system provided the most adequate

metrics for classiﬁcation models. Analyzing the re-

Development of a Context-Free Data Ingestion Mechanism for AutoML

585

sults, we can conclude that the feature selection based

on the context ﬁle produced satisfactory values. All

models performed very well, with high accuracy,

AUC, precision, recall, and F1 score, indicating ro-

bust predictive ability.

4.3 Usability Assessment

An opinion survey was conducted through a form to

assess the user experience when using the Data Min-

ing Framework, speciﬁcally the data ingestion func-

tionality. The questionnaire questions were based on

the AttrakDiff method proposed by (Hassenzahl et al.,

2000), commonly used in academia to validate us-

ability and system quality aspects from the user’s per-

spective. The survey consists of 28 pairs of opposing

words that are used to describe the system in question.

Respondents are required to choose the most appro-

priate description of the system on a scale from -3 to

The study was conducted with 20 technology pro-

fessionals with various levels of education, ranging

from undergraduates (incomplete higher education)

to professionals with a master’s degree. This range

of participants’ knowledge levels allowed for the col-

lection of information from both specialists and non-

specialists in the ﬁeld of data science from a sample

of 20 participants.

As shown in Figure 10, users classiﬁed the data

ingestion mechanism into two categories: “desired”

and “task-oriented”. When classiﬁed as ”desired,”

the software likely provides users with a pleasant,

aesthetically appealing, and emotionally satisfying

experience. This suggests that users positively re-

spond to the software’s design, aesthetics, and sensory

experience. Additionally, being classiﬁed as ”task-

oriented,” as presented in Figure 11, implies that the

software is perceived as efﬁcient, functional, and suit-

able for fulﬁlling the speciﬁc tasks for which it was

designed. This demonstrates that users view the soft-

ware as useful, practical, and aligned with their func-

tional needs. This is a favorable position, representing

a positive balance between the pragmatic and hedonic

aspects of the system.

Figure 10: Portfolio of results.

Figure 11: Diagram of average values.

4.4 Expert Opinion Evaluation

Research projects in software engineering often em-

ploy the expert opinion method (Garcia, 2010), as

scientiﬁc development in software engineering has

many peculiarities to evaluate the quality of the de-

veloped product. However, there are still divergent

opinions on using this methodology, as there is no

universally accepted framework to deal with expert

opinions (Ming Li and Smidts, 2003).

The evaluation process comprised four stages in-

spired by those outlined by (Garcia, 2010). The ﬁrst

stage involved selecting experts based on credibility,

technical knowledge, and availability. Next, opin-

ion elicitation took place by introducing the system

to the experts and providing a questionnaire for them

to describe their considerations. The third stage in-

volved opinion aggregation, where the overall con-

ICEIS 2025 - 27th International Conference on Enterprise Information Systems

586

sensus regarding the obtained responses was assessed.

Finally, results analysis was conducted by examining

the questionnaire responses.

Analyzing the questionnaire responses provides

valuable insights into the relevance of the developed

platform and the research area. Based on expert opin-

ions, the functionalities were successfully validated,

although the tool has room for improvement in fu-

ture developments. The highlighted beneﬁts include

greater convenience in the ingestion process, a shorter

learning curve compared to similar platforms, and

the potential for use in different contexts. Addition-

ally, it was emphasized that the framework is suit-

able for meeting the needs of both technical and non-

technical users. As for improvements, suggestions in-

clude adding new inputs for structured and unstruc-

tured data and options for responsiveness, accessibil-

ity, and internationalization of the platform.

5 CONCLUSION

This work presented the development of a data in-

gestion mechanism customized for AutoML systems,

considering their peculiarities and those of their users.

The project placed a strong emphasis on the personas

identiﬁed during the Lean Inception process, guiding

the entire development process. For this purpose, the

data ingestion mechanism was created and integrated

with an AutoML, named the ”Data Mining Frame-

work.” This mechanism enables data input through a

simple interface, from generic CSV ﬁles and database

connections, mapping data contexts.

In terms of usability, various techniques related to

the lean inception process were applied to optimize

the development of the data ingestion, ensuring that

users can perform this machine learning step with

a low learning curve. To validate the ﬁndings and

ensure research reliability, this aspect was assessed

through an opinion survey with computer engineer-

ing students, along with the expert opinion elicitation

process. The survey results conﬁrmed the relevance

and effectiveness of the project’s identiﬁed function-

alities, providing a good user experience with a lower

learning curve compared to other tools used for the

same purpose and the potential for use in different

contexts.

Regarding computational intelligence, the project

contributes with the development of a data inges-

tion module for automated machine learning systems,

aiming at democratizing data science. The platform

allows storing and transforming the tacit knowledge

of business area experts into explicit knowledge, as-

sisting in the engineering and selection of ideal at-

tributes for applying machine learning techniques in

a speciﬁc context. It is essential to highlight the stan-

dardization of the data ingestion process and the pre-

sentation of an optimized way to input data into the

machine learning workﬂow.

In summary, from a technical point of view, the

proposed data ingestion mechanism introduces inno-

vations by automating and simplifying the prepro-

cessing phase, which is often a bottleneck in Au-

toML workﬂows. Unlike traditional ingestion tools

that require extensive manual intervention, the de-

veloped module allows users to access and lever-

age previously conﬁgured domain knowledge within

the platform, enabling more informed data prepro-

cessing and feature selection. This approach en-

hances the reproducibility of data pipelines by embed-

ding provenance tracking and validation mechanisms,

ensuring consistency in model training. These ad-

vancements contribute to reducing the effort required

from users, making AutoML adoption more accessi-

ble while maintaining data integrity and reliability.

6 FUTURE WORK

In terms of future work, several areas stand out for en-

hancing the Data Ingestor. Based on expert opinions,

the following future improvements have been identi-

ﬁed:

• Expand the data ingestion capabilities to new for-

mats of structured data, such as other database

management systems (DBMS) or ﬁle formats like

XML, JSON, and XLS;

• Expand the data ingestion capabilities to unstruc-

tured data, such as images and PDF ﬁles;

• Add regionalization functionalities that will en-

able international contributions to the tool’s de-

velopment, as it is open source;

• Include accessibility functionalities, allowing us-

age by a broader range of users;

ACKNOWLEDGEMENTS

This paper was ﬁnanced in part by the Coordenac¸

de Aperfeic¸oamento de Pessoal de N

ıvel Superior -

Brazil (CAPES) - Finance Code 001, Fundac¸

ao de

Amparo a Ci

encia e Tecnologia do Estado de Pernam-

buco (FACEPE), the Conselho Nacional de Desen-

volvimento Cient

ıﬁco e Tecnol

ogico (CNPq) - Brazil-

ian research agencies.

Development of a Context-Free Data Ingestion Mechanism for AutoML

587

REFERENCES

Aken, J. E. (2004). Management research based on the

paradigm of the design sciences: The quest for ﬁeld-

tested and grounded technological rules. Journal of

Management Studies, 41(2):219–246.

Alves, G. M. R. and Maciel, A. M. A. (2023). Survey on

Data Ingestion for AutoML (S). In Proceedings of

the International Conference on Software Engineering

and Knowledge Engineering, pages 411–414.

Caroli, P. (2018). Lean Inception: como alinhar pessoas e

construir o produto certo. Editora Caroli.

Chen, Y.-W., Song, Q., and Hu, X. (2021). Techniques for

Automated Machine Learning. ACM SIGKDD Explo-

rations Newsletter, 22(2):35–50.

da Silva, R. (2020). Desenvolvimento de uma Soluc¸

de Aprendizado de M

aquina Automatizado Integr

avel

a M

ultiplos Ambientes Virtuais de Aprendizagem.

Master’s thesis, Universidade de Pernambuco, Recife.

Elshawi, R., Maher, M., and Sakr, S. (2019). Automated

Machine Learning: State-of-The-Art and Open Chal-

lenges. arXiv:1906.02287 [cs, stat].

Garcia, V. C. (2010). RiSE reference model for software

reuse adoption in brazilian companies.

Gonc¸alves, A. F., Maciel, A. M., and Rodrigues, R. L.

(2017). Development of a data mining education

framework for visualization of data in distance learn-

ing environments. International Conferences on Soft-

ware Engineering and Knowledge Engineering.

H. R. Mac

edo, P., B. Santos, W., and M. A. Maciel, A.

(2021). An

alise de Perﬁs de Engajamento de Estu-

dantes de Ensino a Dist

ancia. RENOTE, 18(2):326–

335.

Hapke, H. and Nelson, C. (2020). Building Machine Learn-

ing Pipelines: Automating model life cycles with ten-

sorﬂow. O’Reilly Media, Sebastopol, CA.

Hassenzahl, M., Platz, A., Burmester, M., and Lehner, K.

(2000). Hedonic and ergonomic quality aspects de-

termine a software’s appeal. In Proceedings of the

SIGCHI conference on Human Factors in Computing

Systems, pages 201–208, The Hague The Netherlands.

ACM.

Hlupic, T. and Punis, J. (2021). An Overview of Cur-

rent Trends in Data Ingestion and Integration. In

2021 44th International Convention on Information,

Communication and Electronic Technology (MIPRO),

pages 1265–1270, Opatija, Croatia. IEEE.

Hutter, F., Kotthoff, L., and Vanschoren, J., editors (2019).

Automated Machine Learning: Methods, Systems,

Challenges. The Springer Series on Challenges in

Machine Learning. Springer International Publishing,

Cham.

Ming Li and Smidts, C. (2003). A ranking of software en-

gineering measures based on expert opinion. IEEE

Transactions on Software Engineering, 29(9):811–

824.

Nagarajah, T. and Poravi, G. (2019). A Review on Au-

tomated Machine Learning (AutoML) Systems. In

2019 IEEE 5th International Conference for Conver-

gence in Technology (I2CT), pages 1–6, Bombay, In-

dia. IEEE.

Patel, J. (2020). The Democratization of Machine Learning

Features. In 2020 IEEE 21st International Confer-

ence on Information Reuse and Integration for Data

Science (IRI), pages 136–141, Las Vegas, NV, USA.

IEEE.

Ray, S., AlGhamdi, A., AlGhamdi, A., Alshouiliy, K., and

Agrawal, D. P. (2020). Selecting Features for Breast

Cancer Analysis and Prediction. In 2020 International

Conference on Advances in Computing and Communi-

cation Engineering (ICACCE), pages 1–6, Las Vegas,

NV, USA. IEEE.

Saurav, S. and Schwarz, P. (2016). A Machine-Learning

Approach to Automatic Detection of Delimiters in

Tabular Data Files. In 2016 IEEE 18th Inter-

national Conference on High Performance Com-

puting and Communications; IEEE 14th Interna-

tional Conference on Smart City; IEEE 2nd Inter-

national Conference on Data Science and Systems

(HPCC/SmartCity/DSS), pages 1501–1503, Sydney,

Australia. IEEE.

Tidwell, J., Brewer, C., and Valencia, A. (2020). Designing

interfaces: Patterns for effective interaction design.

O’Reilly Media, Sebastopol, CA.

Whalen, J. (2019). Design for How People Think: Using

Brain Science to Build Better Products. O’Reilly Me-

dia, Sebastopol, CA.

oller, M.-A. and Huber, M. F. (2021). Benchmark and

Survey of Automated Machine Learning Frameworks.

Journal of Artiﬁcial Intelligence Research, 70:409–

472.

ICEIS 2025 - 27th International Conference on Enterprise Information Systems

588