Towards AI-Enabled Model-Driven Architecture:

Systematic Literature Review

Zina Zammel, Mouna Rekik, Lotfi Souifi and Ismael Bouassida Rodriguez

ReDCAD Laboratory, ENIS, University of Sfax, Tunisia

Keywords: Artificial Intelligence, Systematic Literature Review, Model Driven Architecture, CIM, PIM, PSM,

Code Generation.

Abstract: The convergence of two separate areas of computer science, like Model Driven Architecture and Artificial

Intelligence, can lead to collaboration in two main ways, such as AI-driven MDA and MDA for AI. In this

paper, we present a Systematic Literature Review (SLR) on the application of AI within MDA. Additionally,

we examine how AI facilitates transformations between the Computation Independent Model (CIM), Platform

Independent Model (PIM), and Platform Specific Model (PSM), highlighting methods that bridge conceptual

models with technical specifications. This review contributes to a deeper understanding of AI’s role in enhanc-

ing the effectiveness of MDA frameworks by analyzing existing studies that are selected using SLR. Based

on a systematic search of IEEE, Science Direct, Springer, ACM, and google scholarrelevant articles published

between 2018 and 2024 were identified. The adoption of AI introduces numerous benefits to software engi-

neering, including enhanced support for designers and automation in model transformations.

1 INTRODUCTION

Implementing a software solution to meet business

needs is a complex process that involves several steps.

The first step is to translate the stakeholders’ needs

into the requirements of the future system, usually

in a specification written in natural language. The

Model Driven Architecture (MDA) approach plays

an important role in addressing software complexity,

ensuring consistency between different levels of sys-

tem design, and facilitating application maintenance

and evolution of applications (Zouani and Lachgar,

2024). Business Process Modeling Notation (BPMN)

and UML are universally accepted standards for de-

signing models in software development process us-

ing MDA.

MDA advocated by the Object Management

Group (OMG) to highlight the importance of

abstract mod- eling (Zouani and Lachgar, 2024).

MDA defines three primary types of models : CIM

represents the system’s requirements and business

context without describing the structure or

processing ; PIM spec- ifies the structure and

functionality of the system namely, models abstracts

the details of PSM that provides technical

information on the implementa- tion of the system

using a particular technology or platform. In MDA,

the requirements specified in a CIM must be

traceable to the constructs in the PIM and the PSMs

that implement them. Further- more, MDA facilitates

a model-driven software de- velopment process

through Model-to-Model (M2M) transformations,

CIM requirements are transformed into PIM, which

focuses on software functionality rather than

implementation details. A Model-to-Text (M2T)

transformation converts PIM models into PSM or

source code. Acceleo, Xtend, EGL, TextGen, and

AdoScript are common M2T languages, while ATL,

EGL, and QVT-Operational are commonly used for

M2M transformations. Utilizing MDA signifi- cantly

lowers software development costs compared to

conventional Software Development Life Cycle ap-

proaches, while maintaining high quality since code

is generated from the established models. Moreover,

Domain-Specific Models (DSM) enhance MDA) by

enabling the creation of abstract, domain-focused rep-

resentations that capture system requirements and de-

signs across various domains. DSM utilizes two

modeling notations: graphical, as in Domain-Specific

Modeling Languages (DSML), and textual, as in

Domain-Specific Languages (DSL). Despite its ben-

efits, DSM faces challenges such as domain-specific

customization, maintaining consistency across ab-

1380

Zammel, Z., Rekik, M., Souiﬁ, L. and Rodriguez, I. B.

Towards AI-Enabled Model-Driven Architecture: Systematic Literature Review.

DOI: 10.5220/0013374000003890

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

In Proceedings of the 17th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2025) - Volume 3, pages 1380-1387

ISBN: 978-989-758-737-5; ISSN: 2184-433X

straction levels, and managing transformations into

platform-specific implementations. Semi-formal

languages like UML, BPMN, and DSLs are

commonly used to define PIMs, offering structured

notations but lacking the precision of formal

languages. This can result in ambiguities, unclear

semantics, and overlap- ping interpretations,

particularly in complex scenar- ios. On the other

hand, extracting PIM involves con- verting system

requirements, which are often written in natural

language, into formal or semi-formal rep-

resentations. This task has traditionally been the re-

sponsibility of humans due to the inherent complexity

and ambiguity of natural language, making it chal-

lenging for machines to process. However, recent

advancement in AI algorithms have shown promise

in addressing these challenges by automating aspects

of requirement analysis, Natural Language Process-

ing (NLP), and model generation. Several system-

atic literature reviews on Model Driven Architecture

have been conducted, such as the work by (Uzun and

Tekinerdogan, 2018), which examined various Model

Driven Architecture Based Testing approaches. Their

SLR revealed that although MDABT is a generic pro-

cess, the available approaches differed in their spe-

cific goals, modeling abstractions, and results. To the

best of our knowledge, there is currently no existing

SLR that focuses on the application of artificial intel-

ligence within the context of MDA.

In this review, we explore how AI techniques have

been applied to enhance the MDA process, specif-

ically in overcoming challenges such as ambiguity

of the requirements, consistency of the model, and

automated transformation. In addition, we identify

and discuss the limitations and opportunities these

approaches present. This synthesis aims to provide

researchers with insight to select appropriate AI al-

gorithms, address persistent challenges, and explore

future directions to advance the field of MDA. The

remainder of this paper is structured as follows. In

Section 2, we describe the adopted methodology for

conducting our SLR. We describe the search strategy,

the inclusion and exclusion criteria, and the data ex-

traction process. Section 3 presents the results of our

SLR and provides an analysis of the identified stud-

ies. In Section 4, we discuss the limitations of some

proposed studies. In section 5, we present threats to

validity of our SLR. In Section 6 we present a sum-

mary of our paper and our future direction.

2 RESEARCH METHOD

The research methodology for this study follows the

SLR approach, comprising five steps: (i) Defining

research objectives and questions, (ii) Conducting a

literature search with targeted queries, (iii) Selecting

studies using inclusion and exclusion criteria, (iv) Ex-

tracting data from selected studies, and (v) Analyzing

results to address the research questions.

2.1 Research Questions

Identifying specific and valid research questions is the

first step in SLR. To achieve the goal of this work,

we aim to answer the following research questions

(RQs):

•

RQ1 While using MDA what kind of AI algo-

rithm can enhance designer work?

•

RQ2 How are AI algorithms used for CIM gener-

ation?

•

RQ3 How AI algorithms are used for PIM gener-

ation?

•

RQ4 How AI algorithms are used for PSM gener-

ation?

•

RQ5 How AI algorithms are used for Code gen-

eration?

2.2 Search String

A database search strategy was employed to col- lect

relevant published literature, using systematic

searches with well-defined search strings. The re-

search string comprised keywords organized into

three groups.

Group1: “model driven Architecture”.

Group 2: “CIM”, “PIM”, PSM”, “code genera- tion”

Group 3: “Artificial intelligence”.

Both sets of keywords were combined with a Boolean

search (AND, OR) in the article search process. The

search string is reported below :

(“Model Driven Architecture”) AND (“PSM”

OR “CIM” OR ”CODE GENERATION” OR

”PIM ”)

AND (“Artificial intelligence”)

2.3 Selection Criteria

Exclusion Criteria: For our SLR we excluded the

Review papers (survey, SLR . . .), book chap- ters,

master’s, and Ph.D. are excluded. In addi- tion,

publications that were published before or on

31.12.2017, and articles that were written in any

language other than English are excluded.

Inclusion Criteria: Only peer-reviewed studies

published in journals or conference proceedings were

Towards AI-Enabled Model-Driven Architecture: Systematic Literature Review

1381

included. Also, we include the papers that

contributed to solving MDA/ MDE challenges with

AI in the abstract.

2.4 Data Extraction

The search string was used to collect the studies that

are present in multiple sources. Specifically, the

sources considered were IEEE, ACM, Science Direct,

Springer and google scholar. The execution of the de-

fined research query has selected 1261 articles to ob-

tain 52 relevant studies following the application of

the selection criteria. So, the outcomes of the selec-

tion process. 52 articles that matched the inclusion

criteria were included in the SLR. we also included an

additional 4 studies recommended by the expert. The

distribution of selected studies according to the scien-

tific databases and the publication type is presented in

Table 1.

3 RESULTS AND ANALYSIS

This section presents the comprehensive results

obtained from the conducted SLR. The results are

organized and presented according to predefined in-

clusion and exclusion criteria, ensuring the relevance

and quality of the selected studies.

3.1 RQ1: While Using MDA What Kind

of AI Algorithm Can Enhance

Designer Work?

(López et al., 2022) introduced ModelSet, a la- beled

dataset of 5,466 Ecore meta-models and 5,120 UML

models designed to advance machine learning in

MDE. After removing non-English and uncatego-

rized models, the dataset included 5,290 Ecore (14%

”dummy”) and 4,479 UML models (13% ”dummy”).

Features with near-zero variance were eliminated,

leaving 9 features for Ecore and 39 for UML. To ad-

dress class imbalance, upsampling was applied, and

10-fold cross-validation with three repetitions opti-

mized hyperparameters for classifiers, including k-

NN, Random Forest, Neural Networks, and C5.0.

Model performance was evaluated using paired t-

tests for statistical significance. This work estab-

lishes ModelSet as a critical resource for AI-enhanced

MDA, enabling tasks like model classification, tag-

ging, and quality filtering while providing insights

into feature relevance and classifier performance.

(Iyenghar et al., 2022) proposed integrating con-

versational AI frameworks, such as RASA, with

MDE tools to facilitate guided tutorials, natural lan-

guage query resolution, and dialog-based modeling

support. They emphasized using AI techniques like

machine learning for model transformation, semantic

reasoning for DSM, and NLP to simplify modeling

tasks, ultimately reducing complexity and the learn-

ing curve for MDA tools.

(Muttillo et al., 2024) proposed A novel MDE

frame- work integrates event logs, intelligent

modeling as- sistants (IMAs), and LLM generated

modeling oper- ations to automate tasks and provide

recommenda- tions. LLMs generate synthetic data to

train IMAs, though human-based operations are more

accurate. Deep learning techniques like LSTM

networks pre- dict future modeling actions, fostering

proactive de- sign. The authors evaluated the

proposed framework in terms of correctness,

diversity, and hallucination, showing that LLMs,

particularly GPT-4, can effec- tively emulate human

modeling operations. Based on the analysis, GPT-4

outperformed other LLMs across multiple metrics. It

achieved a confidence interval (CI) entirely below 1,

with lower bounds at 0.874 and upper bounds at 0.95,

and an interquartile range (IQR) of 0, indicating

minimal hallucination effects up to the 95th

percentile. Additionally, GPT-4 exhib- ited the lowest

standard error (0.0192), standard de- viation (0.352),

and variance (0.124) among the com- pared models.

A One-Sample Test with a null hy- pothesis that

GPT-4’s mean is greater than 1 yielded a p-value <

0.001, confirming GPT-4’s superior per- formance in

minimizing hallucinations relative to the other

models.

(Bruneliere et al., 2022) proposed innovative com-

bination of MDE, NLP , and DevOps practices. This

integration is intended to facilitate a smoother transi-

tion from design to runtime, thereby improving the

overall efficiency of engineering processes in CPS.

The role NLP is enhances requirement management

by automating writing, ensuring consistency, aiding

in elicitation, improving communication, and manag-

ing the lifecycle of requirements.

(Moin et al., 2022) presented a model-driven soft-

ware engineering methodology that integrates DSM

to enhance the development of AI-enabled IoT sys-

tems. It addresses challenges in traditional MDA,

such as round-tripping and genericity, by focus-

ing on deploying compact ML models on resource-

constrained devices TinyML. This approach allows

for local data processing, improving performance,

availability and privacy. AI algorithms optimized

for low-power environments, like decision trees and

lightweight neural networks, can automate decision-

making and enhance predictive capabilities.

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

1382

Table 1: Summary of selected research studies and publication type.

Scientific database Type Studies Total

IEEE

Journal Not found

Conference

(Kulkarni et al., 2023), (Rigou et al., 2020) ,(Bhadra

et al., 2024),(Lano and Xue, 2023),(Siala, 2024),(Zen- naro et al.,

2018), (Thota et al., 2024), (Binder et al., 2021), (Benaben et al.,

2019),(Tinnes et al., 2021), (Dorodnykh et al., 2018), (Houghtaling et

al., 2024), (Moin et al., 2022), (Babaalla et al., 2024a), (Park et al.,

2023),

ACM

Journal Not found

Conference

(Yang and Sahraoui, 2022), (Safdar et al., 2022), (Kouissi

et al., 2019), (Chang et al., 2020), (Babaalla et al., 2024b), (Uyanık

and Sayar, 2023), (Sajji et al., 2023)

Springer

Journal

(Binder et al., 2022),(Mythily et al., 2019), (Panahandeh

et al., 2021) , (Ouali et al., 2020), (Biswas et al., 2022), (Li et al., 2018),

(Pe´rez-Castillo et al., 2022)

Conference Not found

Science Direct Journal

(Batchkova and Ivanova, 2019), (Maass and Storey,

2021), (Alulema et al., 2023), (Servadei et al., 2019)

Google Scholar

Journal

(Zouani and Lachgar, 2024),(Brandon et al., 2024) ,

(Bruneliere et al., 2022), (Eramo et al., 2024), (Khalfi et al.,

2024),(Lo´pez et al., 2022), (Moin et al., 2021) , (Ouchra et al., 2024),

(Tabbiche et al., 2023)

Conference

, (Naveed et al., 2024), (Koseler et al., 2019) , (Sarazin

et al., 2021), (Naimi et al., 2024), (Meyma et al., 2022), (Lopes et al.,

2024), (Iyenghar et al., 2022), (Liu et al., 2020), (Muttillo et al., 2024)

(Bran- don et al., 2024) implemented CINCO de

Bio, a low-code platform that simplifies biomedical

imaging workflows by integrating model-driven

architecture and AI. It enables non-technical users

to design and execute computational workflows,

offering modular- ity, scalability, and semantic

validation. (Park et al., 2023) used MDA to

facilitates the automation and ab- straction of the

design process for neuromorphic ar- chitectures,

enabling efficient exploration of hetero- geneous

configurations and multi-objective optimiza- tion.

AI plays a crucial role in guiding the design space

search, allowing for the identification of op- timal

architectural candidates based on performance

metrics and user-defined constraints. (Eramo et al.,

2024) proposed a novel architecture MDE, AI/ML,

and DevOps automates processes like requirements,

modeling, coding, testing, and monitoring. It lever-

ages NLP for requirements analysis and GNNs for

modeling insights, while AI/ML supports code gener-

ation and reuse recommendations, enhancing system

engineering and continuous delivery.

(Kulkarni et al., 2023) proposed a use case

demon-strated how Generative AI, specifically

ChatGPT, en- hances the MDE process by enabling

domain experts to create models using natural

language. Using in- puts such as a root goal, a meta-

model description, and context, ChatGPT generates

actionable strategies, like improving academic

reputation and research out-put, as model instances.

This approach bridges human intentions with technical

models, streamlining MDE through natural language

interactions while ensuring traceability and accuracy.

RQ3: How AI algorithms are used for PIM

generation?

(Siala, 2024) Developed a model-driven reverse

engi- neering approach tool using LLMs to generate

UML and OCL specifications from source code

because legacy systems become increasingly

complex and more difficult to maintain, there is a

growing need for new and better ways to understand

and maintain them. Also, Class diagrams, which

visually repre- sent classes, attributes, methods, and

their interrela- tionships, play a pivotal role in

capturing the core ar- chitecture of a software system.

To address this need, (Sajji et al., 2023) proposed an

approach that employs Graph Neural Networks, to

automatically generate class diagrams from source

code in the context of MDA and reverse engineering.

According to the pa- per, software development

company faces challenges with a complex,

undocumented codebase, leading to longer

development cycles and increased energy con-

sumption. To address this, a Graph Neural Network

(GNN) is used to generate class diagrams from the

source code, improving system understanding. Fo-

Towards AI-Enabled Model-Driven Architecture: Systematic Literature Review

1383

cusing on a school management system in Java, the

source code is analyzed to identify classes, attributes,

methods, and relationships like inheritance and de-

pendencies. A graph representation is constructed,

with nodes for classes and edges for relationships,

while relevant features are extracted. Trained on la-

beled datasets of code graphs and class diagrams, the

GNN accurately captures class relationships and gen-

erates diagrams that enhance code clarity, streamline

workflows, and reduce energy consumption.

Manually creating UML and use case diagrams

can be tedious and error-prone, especially when the

specifications are long and/or complex. As a result,

(Babaalla et al., 2024a) suggested a novel method

for examining textual specifications and extracting the

relevant elements needed to create UML class and

use case diagrams, utilizing NLP tools and linguis-

tic techniques. Knowledge extraction module used

for generating the concepts of the two UML diagrams

of classes and use cases, from the output of the NLP

module process used to analyze the text requirements.

The elements of the two resulting diagrams are cre-

ated using drawing algorithms and/or saved in XML

format. The results demonstrated a high F1 score for

the extraction of classes, attributes, and methods was

reported to be in the interval of [80%; 100%]. In

the same way (Yang and Sahraoui, 2022) presented a

novel automated approach for generating UML class

diagrams from natural language specifications.To de-

velop this approach, they created a dataset of UML

class diagrams and their English specifications. The

pipeline of this work included several steps: segment-

ing the input text into sentences, classifying these

sentences, generating UML class diagram fragments,

and composing these fragments into a complete UML

class diagram using natural language patterns and ma-

chine learning. They used Bernoulli Naive Bayes

classifier binary classification of English sentences.

The evaluation metrics results for classes are 17%

precision and 25% recall for exact matching, the

strictest metric. The results for relationships are a

connectivity similarity of 63% and a size difference

of 67%. (Tinnes et al., 2021) proposed OCKHAM,

an unsupervised approach that learns domain-specific

edit operations from model histories in repositories

using frequent subgraph mining to identify meaning-

ful patterns in model differences. The approach was

evaluated on synthetic EMF models and a large-scale

railway case study, demonstrating its ability to extract

and recommend relevant edit operations in real-world

settings. Furthrmore, (Rigou et al., 2020) Analyzed

machine learning approaches for draft a PIM model

that describes the functional requirements of a system

from a textual specification.

(Tabbiche et al., 2023) proposed an intelligent

meta- model that integrates the Eclipse Modeling

Frame- work (EMF) with supervised machine

learning to en- hance the adaptability and efficiency

of the modeling process, particularly for ubiquitous

applications. The approach employs multi-layer

Perceptron (MLP) neu- ral networks to classify and

predict outcomes based on contextual data, such as

COVID-19 symptoms. A systematic process for PIM

generation involves defin- ing a context meta-model

and creating PIMs that are platform-independent.

Using Acceleo for M2T trans- formation automates

code generation for specific plat- forms, ensuring the

relevance and adaptability of the generated models.

The use case focuses on improv- ing COVID-19

patient classification using symptoms as input to an

MLP-based neural network with one or two hidden

layers. A dataset from the COVID- 19-TRACERSET

repository by Public Health France serves as the

training and test data. Initial KNN weights are

randomly set and adjusted during learn- ing, where

outputs above a threshold (> 0.5) are la- beled as

“infected” (1) or “healthy” (0). Experimen- tal

results demonstrate the reliability of ANN models in

automatic decision-making, showcasing their accu-

racy in categorizing COVID-19 cases based on symp-

toms, ultimately improving the decision-making pro-

cess.

3.2 RQ5 How AI Algorithms Are Used

for Code Generation?

(Lano and Xue, 2023) applied novel symbolic ma-

chine learning techniques for learning tree-to-tree

mappings of software syntax trees, to automate the

development of code generators from source–target

example pairs. This approach is referred to as Code

Generation by Example. The method was evaluated

across various tasks, including translating UML/OCL

to programming languages such as Java, Kotlin, and

C, as well as translating DSLs to SwiftUI. The re-

sults demonstrate both high accuracy and efficiency.

(Bhadra et al., 2024) introduced a novel model-based

code generator based on MDA to tackle the com-

plexities of embedded programming, highlighting the

need for diverse coding styles due to varying pro-

gramming languages and hardware architectures. The

authors demonstrate that their approach reduces the

effort for generating low-level driver code for core

tensor math operators in neural networks by an av-

erage of 62 times in Source Lines of Code compared

to manual coding. This efficiency enhances the re-

liability and scalability of embedded software solu-

tions, showcasing the value of model-driven method-

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

1384

ologies in automating code generation. While ac-

knowledging the potential of LLMs for code gener-

ation, the authors emphasize the deterministic out-

comes of their approach, which ensures reliability and

optimized code for embedded systems. Overall, the

findings contribute significantly to software engineer-

ing and embedded systems development.

4 LIMITATIONS

AI has shown significant potential to enhance soft-

ware engineering processes, but its application in

MDA is still emerging and faces several limitations.

A major challenge is the lack of large, high-quality

datasets tailored to specific modeling languages, of-

ten due to privacy concerns. While LLMs can gener-

ate synthetic traces to mitigate this issue, the quality

of such data remains critical. Generated traces may

suffer from inaccuracies or ’hallucinations’ outputs

where AI systems generate plausible but incorrect or

nonsensical information (Muttillo et al., 2024) devi-

ating from correct modeling practices and compro-

mising the reliability of AI-driven tools. Addition-

ally, many AI-based approaches lack generalizabil-

ity across different modeling tools and environments.

Validation is often limited, with some studies rely-

ing on a single case study such as a hydraulic test rig

(Moin et al., 2022) which may not represent broader

domains like IoT. This highlights the need for more

comprehensive evaluations across diverse scenarios

and use cases. Finally, the use of LLMs in genera-

tive AI raises ethical and privacy concerns, particu-

larly when handling sensitive or proprietary modeling

data.

5 THREATS TO VALIDITY

To minimize possible threats to validity, particularly

threats to internal and construct validity, we followed

well-established guidelines for studies during this re-

search. In SLR, one of the main threats to external

validity is that primary studies may not be represen-

tative of the state of the art and practice. To miti-

gate this threat, we targeted four well-known scien-

tific databases. These operation also helped us mit-

igate threats to construct validity. However, we re-

moved studies that were not written in English, but

we don’t believe there’s a significant risk of exclud-

ing relevant studies not written in English since

English is the de-facto standard language for

scientific research in computer science and software

engineer- ing. It is possible that SLR cannot answer all

research questions.

6 CONCLUSIONS

In our systematic literature review, we have exam-

ined a total of 56 primary studies to explore the use

of various AI algorithms in MDA. The aim of our re-

view was to address specific research questions and

provide an overview of current MDA that is enabled

by AI. From this SLR, we conclude that, in recent

years, major players have increasingly used model-

based technologies to develop industrial software. AI

has also seen significant advances recently, especially

in Large Language Models.

This paper offers valuable insights for both

practition- ers and researchers examining the current

state of the field. Furthermore, our SLR can

positively influence the research community and

facilitate its transition to Generative AI.

In future works, we plan to propose an intelli- gent

MDA framework that automates and enhances the

generation of requirement, models and the code

generation. Also, We aim to utilize a large datasets to

prevent underfitting during the training of the AI

models.

ACKNOWLEDGEMENTS

This work was partially supported by the LABEX-TA

project MeFoGL: ”Me´thodes formelles pour le Ge´nie

Logiciel”

REFERENCES

Alulema, D., Criado, J., Iribarne, L., Ferna´ndez-

Garc´ıa,A. J., and Ayala, R. (2023). Si4iot: A

methodology based on models and services for the

integration of iot systems. Future Generation

Computer Systems, 143:132–151.

Babaalla, Z., Bouziane, E. M., Jakimi, A., and Oualla, M.

(2024a). From text-based system specifications to uml

diagrams: A bridge between words and models. In 2024

International Conference on Circuit, Systems and

Communication (ICCSC), pages 1–6.

Babaalla, Z., Jakimi, A., Oualla, M., Saadane, R., and

Chehri, A. (2024b). Towards an automatic extracting

uml class diagram from system’s textual specification. In

Proceedings of the 7th International Conference on

Networking, Intelligent Systems and Security, pages 1–

5. Batchkova, I. and Ivanova, T. (2019). Model-driven

Towards AI-Enabled Model-Driven Architecture: Systematic Literature Review

1385

devel- opment of agent-based cyber-physical systems.

IFAC- PapersOnLine, 52(25):258–263.

Benaben, F., Lauras, M., Fertier, A., and Salatge´, N. (2019).

Integrating model-driven engineering as the next chal-

lenge for artificial intelligence – application to risk and

crisis management. In 2019 Winter Simulation

Conference (WSC), pages 1549–1563.

Bhadra, M., Lopera, D. S., Kunzelmann, R., and Ecker,

W. (2024). A model-driven architecture approach to

accelerate software code generation. In 2024 7th In-

ternational Conference on Software and System Engi-

neering (ICoSSE), pages 23–30. IEEE.

Binder, C., Cala`, A., Vollmar, J., Neureiter, C., and Lu¨der,

A. (2021). Automated model transformation in mod-

eling digital twins of industrial internet-of-things ap-

plications utilizing automationml. In 2021 26th IEEE

International Conference on Emerging Technologies

and Factory Automation (ETFA ), pages 1–6.

Binder, C., Neureiter, C., and Lu¨der, A. (2022). Towards a

domain-specific information architecture enabling the

investigation and optimization of flexible production

systems by utilizing artificial intelligence. The In-

ternational Journal of Advanced Manufacturing Tech-

nology, 123(1):49–81.

Biswas, N., Mondal, A. S., Kusumastuti, A., Saha, S., and

Mondal, K. C. (2022). Automated credit assessment

framework using etl process and machine learning. In-

novations in Systems and Software Engineering, pages

1–14.

Brandon, C., Boßelmann, S., Singh, A., Ryan, S.,

Schieweck, A., Fennell, E., Steffen, B., and Mar-

garia, T. (2024). Cinco de bio: A low-code platform

for domain-specific workflows for biomedical imag-

ing research. BioMedInformatics, 4(3):1865–1883.

Bruneliere, H., Muttillo, V., Eramo, R., Berardinelli, L.,

Gómez, A., Bagnato, A., Sadovykh, A., and Cicchetti,

A. (2022). Aidoart: Ai-augmented automation for de-

vops, a model-based framework for continuous devel-

opment in cyber–physical systems. Microprocessors

and Microsystems, 94:104672.

Chang, W., Wei, R., Zhao, S., Wellings, A., Woodcock, J.,

and Burns, A. (2020). Development automation of

real-time java: Model-driven transformation and syn-

thesis. ACM Transactions on Embedded Computing

Systems (TECS), 19(5):1–26.

Dorodnykh, N. O., Yurin, A. Y., and Stolbov, A. B. (2018).

Ontology driven development of rule-based expert

systems. In 2018 3rd Russian-Pacific Conference on

Computer Technology and Applications (RPC), pages

1–6.

Eramo, R., Said, B., Oriol, M., Bruneliere, H., and Morales,

S. (2024). An architecture for model-based and intel-

ligent automation in devops. Journal of Systems and

Software, 217:112180.

Houghtaling, M. A., Fiorini, S. R., Fabiano, N.,

Gonc¸alves, P. J. S., Ulgen, O., Haidegger, T.,

Carbonera, J. L., Olszewska, J. I., Page, B., Murahwi,

Z., and Prestes,

E. (2024). Standardizing an ontology for ethically

aligned robotic and autonomous systems. IEEE

Transactions on Systems, Man, and Cybernetics:

Systems, 54(3):1791–1804.

Iyenghar, P., Otte, F., and Pulvermueller, E. (2022). Ai-

guided model-driven embedded software engineering.

In MODELSWARD, pages 395–404.

Khalfi, M. F., Tabbiche, M. N., and Adjoudj, R.

(2024). From programming-to-modeling-to-prompts

smart ubiquitous applications. Journal of Ambient In-

telligence and Smart Environments, (Preprint):1–39.

Koseler, K., McGraw, K., and Stephan, M. (2019). Realiza-

tion of a machine learning domain specific modeling

language: A baseball analytics case study. In MOD-

ELSWARD, pages 13–24.

Kouissi, M., Ghouch, N. E., and En-naimi, E. M. (2019).

New approach for modeling and developing multi-

agent systems based on case based reasoning. In Pro-

ceedings of the 4th International Conference on Smart

City Applications, pages 1–8.

Kulkarni, V., Reddy, S., Barat, S., and Dutta, J. (2023). To-

ward a symbiotic approach leveraging generative ai for

model driven engineering. In 2023 ACM/IEEE 26th

International Conference on Model Driven En-

gineering Languages and Systems (MODELS), pages

184–193. IEEE.

Lano, K. and Xue, Q. (2023). Code generation by exam-

ple using symbolic machine learning. SN Computer

Science, 4(2):170.

Li, X.-S., Tao, X.-P., Song, W., and Dong, K. (2018).

Aocml: A domain-specific language for model-driven

development of activity-oriented context-aware appli-

cations. Journal of Computer Science and Technology,

33:900–917.

Liu, H., Shen, M., Zhu, J., Niu, N., Li, G., and Zhang, L.

(2020). Deep learning based program generation from

requirements text: Are we there yet? IEEE Transac-

tions on Software Engineering, 48(4):1268–1289.

Lopes, R., Arau´jo, J., da Silveira, D. S., and Sardinha, A.

(2024). A systematic approach to derive conceptual

models from bpmn models. In International Sym-

posium on Business Modeling and Software Design,

pages 83–96. Springer.

López, J. A. H., Ca´novas Izquierdo, J. L., and Cuadrado, J.

S. (2022). Modelset: a dataset for machine learning in

model-driven engineering. Software and Systems

Modeling, pages 1–20.

Maass, W. and Storey, V. C. (2021). Pairing conceptual

modeling with machine learning. Data & Knowledge

Engineering, 134:101909.

Meyma, M. M., Laaz, N., and Mbarki, S. (2022). A

new model-based approach for migrating health 2.0 to

health 3.0 applications. In International Confer- ence

on Advanced Intelligent Systems for Sustainable

Development, pages 673–682. Springer.

Moin, A., Challenger, M., Badii, A., and Gu¨nnemann, S.

(2021). Mde4qai: Towards model-driven engineer- ing

for quantum artificial intelligence. arXiv preprint

arXiv:2107.06708.

Moin, A., Challenger, M., Badii, A., and Gu¨nnemann, S.

(2022). Supporting ai engineering on the iot

edgethrough model-driven tinyml. In 2022 IEEE 46th

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

1386

An- nual Computers, Software, and Applications

Confer- ence (COMPSAC), pages 884–893.

Muttillo, V., Di Sipio, C., Rubei, R., Berardinelli, L., and

Dehghani, M. (2024). Towards synthetic trace gener-

ation of modeling operations using in-context learn-

ing approach. In Proceedings of the 39th IEEE/ACM

International Conference on Automated Software En-

gineering, pages 619–630.

Mythily, M., Valarmathi, M., and Durai, C. A. D. (2019).

Model transformation using logical prediction from

sequence diagram: an experimental approach. Clus-

ter Computing, 22(Suppl 5):12351–12362.

Naimi, L., Bouziane, E. M., and Jakimi, A. (2024). Au-

tomating test case generation from class diagram us-

ing generative ai. In International Conference on

Smart Medical, IoT & Artificial Intelligence, pages

133–140. Springer.

Naveed, H., Grundy, J., Arora, C., Khalajzadeh, H., and

Haggag, O. (2024). Towards runtime monitoring for

responsible machine learning using model-driven en-

gineering. In Proceedings of the ACM/IEEE 27th In-

ternational Conference on Model Driven Engineering

Languages and Systems, pages 195–202.

Ouali, S., Mhiri, M., and Gargouri, F. (2020). Existing

business process models for model construction of

multidimensional business knowledge: an mda-based

transformation methodology. New Generation Com-

puting, 38(3):477–508.

Ouchra, H., Belangour, A., Erraissi, A., and Labied, M.

(2024). Data to cartography new mde-based approach

for urban satellite image classification. Journal of En-

vironmental & Earth Sciences— Volume, 7(01).

Panahandeh, M., Hamdaqa, M., Zamani, B., and Hamou-

Lhadj, A. (2021). Muppit: A method for using proper

patterns in model transformations. Software and Sys-

tems Modeling, 20(5):1491–1523.

Park, J., Shin, Y., and Sung, H. (2023). Multi-objective ar-

chitecture search and optimization for heterogeneous

neuromorphic architecture. In 2023 IEEE/ACM Inter-

national Conference on Computer Aided Design (IC-

CAD), pages 1–8.

Pérez-Castillo, R., Delgado, A., Ruiz, F., Bacigalupe, V.,

and Piattini, M. (2022). A method for transforming

knowledge discovery metamodel to archimate models.

Software And Systems Modeling, pages 1–26.

Rigou, Y., Lamontagne, D., and Khriss, I. (2020). A

sketch of a deep learning approach for discovering uml

class diagrams from system’s textual specifica- tion.

In 2020 1st International Conference on Inno- vative

Research in Applied Science, Engineering and

Technology (IRASET), pages 1–6. IEEE.

Safdar, A., Azam, F., Anwar, M. W., Akram, U., and

Rasheed, Y. (2022). Modlf: A model-driven deep

learning framework for autonomous vehicle percep-

tion (avp). In Proceedings of the 25th International

Conference on Model Driven Engineering Languages

and Systems, pages 187–198.

Sajji, A., Rhazali, Y., and Hadi, Y. (2023). A method-

ology of automatic class diagrams generation from

source code using model-driven architecture and ma-

chine learning to achieve energy efficiency. In E3S

Web of Conferences, volume 412, page 01002. EDP

Sciences.

Sarazin, A., Truptil, S., Montarnal, A., Bascans, J., and

Lorca, X. (2021). Model transformation from cbm to

epl rules to detect failure symptoms. In Model- Driven

Engineering and Software Development: 8th

International Conference, MODELSWARD 2020, Val-

letta, Malta, February 25–27, 2020, Revised Selected

Papers 8, pages 200–224. Springer.

Servadei, L., Zennaro, E., Fritz, T., Devarajegowda, K.,

Ecker, W., and Wille, R. (2019). Using machine learn-

ing for predicting area and firmware metrics of hard-

ware designs from abstract specifications. Micropro-

cessors and Microsystems, 71:102853.

Siala, H. A. (2024). Enhancing model-driven reverse en-

gineering using machine learning. In Proceedings of

the 2024 IEEE/ACM 46th International Conference on

Software Engineering: Companion Proceedings,

ICSE-Companion ’24, page 173–175, New York, NY,

USA. Association for Computing Machinery.

Tabbiche, M. N., Khalfi, M. F., and Adjoudj, R. (2023). Ap-

plying machine learning and model-driven approach for

the identification and diagnosis of covid-19. In-

ternational Journal of Distributed Systems and Tech-

nologies (IJDST), 14(1):1–27.

Thota, S. R., Arora, S., and Gupta, S. (2024). Al-driven

automated software documentation generation for en-

hanced development productivity. In 2024 Interna-

tional Conference on Data Science and Network Se-

curity (ICDSNS), pages 1–7.

Tinnes, C., Kehrer, T., Joblin, M., Hohenstein, U., Biesdorf,

A., and Apel, S. (2021). Learning domain-specific edit

operations from model repositories with frequent

subgraph mining. In 2021 36th IEEE/ACM Interna-

tional Conference on Automated Software Engineer-

ing (ASE), pages 930–942.

Uyanık, B. and Sayar, A. (2023). Analysis and comparison

of automatic code generation and transformation tech-

niques on low-code platforms. In Proceedings of the

2023 5th International Conference on Software Engi-

neering and Development, pages 17–27.

Uzun, B. and Tekinerdogan, B. (2018). Model-driven archi-

tecture based testing: A systematic literature review.

Information and Software technology, 102:30–48.

Yang, S. and Sahraoui, H. (2022). Towards automatically

extracting uml class diagrams from natural language

specifications. In Proceedings of the 25th Interna- tional

Conference on Model Driven Engineering Lan- guages

and Systems: Companion Proceedings, pages 396–403.

Zennaro, E., Servadei, L., Devarajegowda, K., and Ecker,

W. (2018). A machine learning approach for area pre-

diction of hardware designs from abstract specifica-

tions. In 2018 21st Euromicro Conference on Digital

System Design (DSD), pages 413–420.

Zouani, Y. and Lachgar, M. (2024). Zynerator: Bridg- ing

model-driven architecture and microservices for

enhanced software development. Electronics,

13(12):2237.

Towards AI-Enabled Model-Driven Architecture: Systematic Literature Review

1387