Maha Khemaja
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Maha Khemaja has a total of 53 co-authors. Here are the top co-authors by number of shared publications:
- Aroua Taamallah
University of Sousse
8 shared publications - Félix Buendía
Universitat Politècnica de València
8 shared publications Sami Faiz
University of Tunis El Manar
6 shared publications- Roua Jabla
University of Sousse
5 shared publications Valérie Monfort
SOIE, ISG Tunis, Tunis, Tunisia
5 shared publications- Amani Braham
University of Sousse, ISITCom
4 shared publications
Latest Publications
- Open Access
- 20 September 2026
- Published by Fuji Technology Press Ltd. in Journal of Advanced Computational Intelligence and Intelligent Informatics
- Vol. 30 (5) , 1620-1639
AI_MDD Approaches for Generating Adaptive and Intelligent UIs: A Systematic Literature Review
This study presents a combined bibliometric analysis and systematic literature review of artificial intelligence (AI)_model-driven development (MDD) approaches for generating adaptive and intelligent user interfaces. Based on an initial set of 3,335 records retrieved from Scopus, a structured filtering process following preferred reporting items for systematic reviews and meta-analyses guidelines resulted in 51 relevant studies for in-depth analysis. The findings indicate a rapid growth in research within this area, particularly over the last five years, with substantial contributions from the fields of computer science and engineering. The analysis shows that current approaches increasingly rely on integrating machine learning techniques within model-driven pipelines to support context-aware adaptation, particularly in emerging environments such as augmented reality, virtual reality, virtual humans, and metaverse environments. This review further identified a set of dominant methodological trends, including AI-assisted code generation, models@runtime adaptation, and generative user interface approaches. However, several limitations remain evident in the literature, particularly regarding real-time adaptation, cross-platform interoperability, and energy efficiency. Overall, the results suggest that while AI_MDD combination has progressed toward more adaptive and intelligent interface generation, existing approaches remain largely experimental and lack standardized frameworks for deployment. These findings highlight the need for robust methodologies that combine model-based abstraction with data-driven adaptation in a consistent and scalable manner.
- 20 September 2026
- Published by Fuji Technology Press Ltd. in Journal of Advanced Computational Intelligence and Intelligent Informatics
- Vol. 30 (5) , 1620-1639
Open AccessAI_MDD Approaches for Generating Adaptive and Intelligent UIs: A Systematic Literature Review
This study presents a combined bibliometric analysis and systematic literature review of artificial intelligence (AI)_model-driven development (MDD) approaches for generating adaptive and intelligent user interfaces. Based on an initial set of 3,335 records retrieved from Scopus, a structured filtering process following preferred reporting items for systematic reviews and meta-analyses guidelines resulted in 51 relevant studies for in-depth analysis. The findings indicate a rapid growth in research within this area, particularly over the last five years, with substantial contributions from the fields of computer science and engineering. The analysis shows that current approaches increasingly rely on integrating machine learning techniques within model-driven pipelines to support context-aware adaptation, particularly in emerging environments such as augmented reality, virtual reality, virtual humans, and metaverse environments. This review further identified a set of dominant methodological trends, including AI-assisted code generation, models@runtime adaptation, and generative user interface approaches. However, several limitations remain evident in the literature, particularly regarding real-time adaptation, cross-platform interoperability, and energy efficiency. Overall, the results suggest that while AI_MDD combination has progressed toward more adaptive and intelligent interface generation, existing approaches remain largely experimental and lack standardized frameworks for deployment. These findings highlight the need for robust methodologies that combine model-based abstraction with data-driven adaptation in a consistent and scalable manner.
- 28 August 2026
- Published by Springer Nature
- p. 267-280
Simulating Smart-City Strategies
- Open Access
- 22 June 2026
- Published by MDPI AG in Preprints.org
Bridging Model-Driven Engineering and Machine Learning for Adaptive User Interface Generation: A Compact Metamodel, a Learned Layout Selector, and a Reproducible Multi-Metric Evaluation
Automatic user-interface (UI) generation is pursued by two communities that rarely meet. Model-driven engineering (MDE) offers rigour, traceability, and reproducibility but is held back by the cognitive cost of elaborate modelling languages; machine learning (ML) offers flexibility and strong results on visual-to-code tasks but produces opaque, hard-to-reproduce systems that falter on structurally irregular inputs. We argue that the two are complementary rather than competing, and present a hybrid framework in which a deterministic, model-driven backbone is augmented by a single, clearly bounded learned component. The framework couples a deliberately compact Platform-Independent Model (PIM)—an Ecore metamodel of seven first-class concepts—with five deterministic layout primitives rendered through templates that consume real model content, a learned selector that chooses among the primitives from interpretable structural features, and a multi-metric quality stack anchored by a pre-registered human-evaluation protocol. The entire pipeline is containerised and released for independent reproduction. On a class-balanced corpus constructed by stratified sampling calibrated to published UI-corpus statistics, a gradient-boosted selector attains 0.927 accuracy and 0.927 macro-F1 on held-out data (five-fold CV 0.919 ± 0.010; top-3 accuracy 1.000), an improvement of 86 percentage points in macro-F1 over a majority-class baseline; its errors concentrate exactly where the layout primitives genuinely resemble one another. A perceptual analysis independently corroborates that error structure and shows that two closely related pixel-based metrics diverge under spatial perturbation (inter-metric Pearson r = 0.965, 95% bootstrap CI [0.944, 0.983]), providing direct empirical support for evaluating UI fidelity with multiple, rather than single, measures. We report candidly which claims the experiments settle and which remain to be confirmed on real-world data at scale. The contribution is a reproducible, accountable, and accessibility-aware route to UI generation that does not force a choice between the rigour of MDE and the flexibility of ML.
- 22 June 2026
- Published by MDPI AG in Preprints.org
Open AccessBridging Model-Driven Engineering and Machine Learning for Adaptive User Interface Generation: A Compact Metamodel, a Learned Layout Selector, and a Reproducible Multi-Metric Evaluation
Automatic user-interface (UI) generation is pursued by two communities that rarely meet. Model-driven engineering (MDE) offers rigour, traceability, and reproducibility but is held back by the cognitive cost of elaborate modelling languages; machine learning (ML) offers flexibility and strong results on visual-to-code tasks but produces opaque, hard-to-reproduce systems that falter on structurally irregular inputs. We argue that the two are complementary rather than competing, and present a hybrid framework in which a deterministic, model-driven backbone is augmented by a single, clearly bounded learned component. The framework couples a deliberately compact Platform-Independent Model (PIM)—an Ecore metamodel of seven first-class concepts—with five deterministic layout primitives rendered through templates that consume real model content, a learned selector that chooses among the primitives from interpretable structural features, and a multi-metric quality stack anchored by a pre-registered human-evaluation protocol. The entire pipeline is containerised and released for independent reproduction. On a class-balanced corpus constructed by stratified sampling calibrated to published UI-corpus statistics, a gradient-boosted selector attains 0.927 accuracy and 0.927 macro-F1 on held-out data (five-fold CV 0.919 ± 0.010; top-3 accuracy 1.000), an improvement of 86 percentage points in macro-F1 over a majority-class baseline; its errors concentrate exactly where the layout primitives genuinely resemble one another. A perceptual analysis independently corroborates that error structure and shows that two closely related pixel-based metrics diverge under spatial perturbation (inter-metric Pearson r = 0.965, 95% bootstrap CI [0.944, 0.983]), providing direct empirical support for evaluating UI fidelity with multiple, rather than single, measures. We report candidly which claims the experiments settle and which remain to be confirmed on real-world data at scale. The contribution is a reproducible, accountable, and accessibility-aware route to UI generation that does not force a choice between the rigour of MDE and the flexibility of ML.
- Open Access
- 13 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 842-846
Behind Agentic Pull Requests: An Empirical Study on Developer Interventions in AI Agent-Authored Pull Requests
- 13 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 842-846
Open AccessBehind Agentic Pull Requests: An Empirical Study on Developer Interventions in AI Agent-Authored Pull Requests
- 7 March 2025
- Published by Springer Nature
- p. 79-99
Design and Evaluation of the UI/UX of GAmified LEarning Design Editor (GALEDE)
- 7 March 2025
- Published by Springer Nature
- p. 79-99
Design and Evaluation of the UI/UX of GAmified LEarning Design Editor (GALEDE)
- Open Access
- 19 October 2022
- Published by Elsevier in Procedia Computer Science
- Vol. 207, 1871-1880
A knowledge-driven activity recognition framework for learning unknown activities
- 19 October 2022
- Published by Elsevier in Procedia Computer Science
- Vol. 207, 1871-1880
Open AccessA knowledge-driven activity recognition framework for learning unknown activities
- 1 September 2022
- Published by Springer Nature
- p. 139-148
Towards a Model-Driven Ontology-Based Architecture for Generating Adaptive User Interfaces
- 1 September 2022
- Published by Springer Nature
- p. 139-148
Towards a Model-Driven Ontology-Based Architecture for Generating Adaptive User Interfaces
- 1 September 2022
- Published by Springer Nature
- p. 127-137
A Novel Component of Decision-Making for Context-Aware Applications in Pervasive Environments
- 1 September 2022
- Published by Springer Nature
- p. 127-137
A Novel Component of Decision-Making for Context-Aware Applications in Pervasive Environments