Luiz Fernando Capretz
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Research Interests
Luiz Fernando Capretz has a total of 228 co-authors. Here are the top co-authors by number of shared publications:
Faheem Ahmed
30 shared publications- Abdullateef Oluwagbemiga Balogun
Department of Computing, Universiti Teknologi PETRONAS, Seri Iskandar, Malaysia
21 shared publications Danny Ho
21 shared publications- Shuib Basri
Department of Computing, Universiti Teknologi PETRONAS, Seri Iskandar, Malaysia
21 shared publications - Ali Bou Nassif
Department of Computer Engineering, University of Sharjah, Sharjah 27272, United Arab Emirates
18 shared publications Arif Raza
Department of Electrical and Computer Engineering, Western University, London, Ontario, Canada
12 shared publications
Latest Publications
- Open Access
- 11 September 2026
- Published by BON VIEW PUBLISHING PTE in Journal of Computational and Cognitive Engineering
An Online Clustering Algorithm for Handling Evolving Data Streams with the Ability to Prevent Clusters' False Merging Using Adaptive Time Interval
Online clustering of evolving data streams presents unique challenges, particularly the problem of false merging, where distinct clusters are incorrectly combined during temporal overlap. Existing buffer-based and density-based algorithms, such as BOCEDS and CEC-Merge, lack mechanisms to adaptively distinguish transient overlaps from genuine merging events. To address this, we propose BOCEDS-ATI—a Buffer-based Online Clustering for Evolving Data Streams with Adaptive Time Interval algorithm. The core innovation is the adaptive time interval mechanism, which dynamically adjusts merging decisions based on the temporal persistence and relative velocity of cluster interactions. This allows the model to recognize and prevent false merging in continuously evolving data streams. Experimental evaluation on multiple synthetic and real-world datasets demonstrates that BOCEDS-ATI achieves superior clustering accuracy and robustness to dynamic drift compared with state-of-the-art algorithms, while maintaining linear time complexity suitable for real-time applications. The results confirm that incorporating localized temporal adaptation significantly improves the reliability of online clustering in non-stationary environments. Received: 17 July 2025 | Revised: 2 December 2025 | Accepted: 6 February 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement Data are available from the corresponding author upon reasonable request. Author Contribution Statement Redhwan Al-amri: Conceptualization, Methodology, Software, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization. Gamal Alkawsi: Validation, Project administration. Abdulnaser A. Hagar: Writing – review & editing. Luiz Fernando Capretz: Supervision, Funding acquisition.
- 11 September 2026
- Published by BON VIEW PUBLISHING PTE in Journal of Computational and Cognitive Engineering
Open AccessAn Online Clustering Algorithm for Handling Evolving Data Streams with the Ability to Prevent Clusters' False Merging Using Adaptive Time Interval
Online clustering of evolving data streams presents unique challenges, particularly the problem of false merging, where distinct clusters are incorrectly combined during temporal overlap. Existing buffer-based and density-based algorithms, such as BOCEDS and CEC-Merge, lack mechanisms to adaptively distinguish transient overlaps from genuine merging events. To address this, we propose BOCEDS-ATI—a Buffer-based Online Clustering for Evolving Data Streams with Adaptive Time Interval algorithm. The core innovation is the adaptive time interval mechanism, which dynamically adjusts merging decisions based on the temporal persistence and relative velocity of cluster interactions. This allows the model to recognize and prevent false merging in continuously evolving data streams. Experimental evaluation on multiple synthetic and real-world datasets demonstrates that BOCEDS-ATI achieves superior clustering accuracy and robustness to dynamic drift compared with state-of-the-art algorithms, while maintaining linear time complexity suitable for real-time applications. The results confirm that incorporating localized temporal adaptation significantly improves the reliability of online clustering in non-stationary environments. Received: 17 July 2025 | Revised: 2 December 2025 | Accepted: 6 February 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement Data are available from the corresponding author upon reasonable request. Author Contribution Statement Redhwan Al-amri: Conceptualization, Methodology, Software, Formal analysis, Investigation, Resources, Data curation, Writing – original draft, Writing – review & editing, Visualization. Gamal Alkawsi: Validation, Project administration. Abdulnaser A. Hagar: Writing – review & editing. Luiz Fernando Capretz: Supervision, Funding acquisition.
- Open Access
Inclusion in Software Development Teams: An Empirical Study
In a software development team, people work together in different roles to carry out the activities of the software development process. Human factors have always been critical in managing software teams. However, they have always been overshadowed by a focus on technology and underlying processes. Recently, there has been greater focus on team diversity, allowing for individual differences among team members, such as age, gender, race and ethnicity, culture, etc. The real challenge of team diversity is to assemble a software development team with a balanced representation of various diverse groups and to foster an inclusive work environment where every team member is engaged and productive, especially those in the minority. The research on inclusiveness in software development teams, which goes hand in hand with diversity, appears limited, and this work outlines its objectives and novelty. This exploratory study investigates correlations between specific workplace inclusion factors and perceived self-efficacy among members of software development teams, using the Mor Barak Inclusion–Exclusion Scale.
Open AccessInclusion in Software Development Teams: An Empirical Study
In a software development team, people work together in different roles to carry out the activities of the software development process. Human factors have always been critical in managing software teams. However, they have always been overshadowed by a focus on technology and underlying processes. Recently, there has been greater focus on team diversity, allowing for individual differences among team members, such as age, gender, race and ethnicity, culture, etc. The real challenge of team diversity is to assemble a software development team with a balanced representation of various diverse groups and to foster an inclusive work environment where every team member is engaged and productive, especially those in the minority. The research on inclusiveness in software development teams, which goes hand in hand with diversity, appears limited, and this work outlines its objectives and novelty. This exploratory study investigates correlations between specific workplace inclusion factors and perceived self-efficacy among members of software development teams, using the Mor Barak Inclusion–Exclusion Scale.
- Open Access
- 4 September 2026
- Published by Elsevier in SSRN Electronic Journal
RAES: Redundancy-Aware Evolutionary Search for Individual Fairness Testing
Context: Individual Fairness Testing (IFT) identifies discriminatory instances (IDIs) where individuals who differ only in protected attributes receive different predictions from a software system. Population-based IFT methods can repeatedly evaluate identical inputs, consuming a fixed testing budget without expanding the tested input set. Objectives: This work introduces RAES (Redundancy-Aware Evolutionary Search), an evolutionary framework that operationalises redundancy reduction as an online design requirement for population-based IFT. Methods: RAES redesigns mutation, crossover, and adaptive mutation-rate adjustment to reduce test redundancy and increase unique IDI detection during search. Unit-step mutation reduces zero-effect perturbations in integer-encoded features, half-uniform crossover recombines only loci on which parents differ, and diversity-driven adaptation raises the mutation rate when population diversity declines. RAES is evaluated on five tabular datasets and four classifier families. Results: RAES detects 29.85%–320.08% more unique IDIs at 1.02×–28.04× higher throughput than the evaluated baselines, with 22.32×–43.32× lower redundancy. In multi-attribute testing, RAES detects 1.49×–7.53× more IDIs with 6.42×–231.50× lower redundancy. Retraining with RAES generated IDIs reduces discrimination from 0.045 to 0.016 while maintaining model accuracy. Conclusion: Controlling redundancy during test generation improves IDI detection and throughput for the tabular classifiers evaluated in this study.
- 4 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessRAES: Redundancy-Aware Evolutionary Search for Individual Fairness Testing
Context: Individual Fairness Testing (IFT) identifies discriminatory instances (IDIs) where individuals who differ only in protected attributes receive different predictions from a software system. Population-based IFT methods can repeatedly evaluate identical inputs, consuming a fixed testing budget without expanding the tested input set. Objectives: This work introduces RAES (Redundancy-Aware Evolutionary Search), an evolutionary framework that operationalises redundancy reduction as an online design requirement for population-based IFT. Methods: RAES redesigns mutation, crossover, and adaptive mutation-rate adjustment to reduce test redundancy and increase unique IDI detection during search. Unit-step mutation reduces zero-effect perturbations in integer-encoded features, half-uniform crossover recombines only loci on which parents differ, and diversity-driven adaptation raises the mutation rate when population diversity declines. RAES is evaluated on five tabular datasets and four classifier families. Results: RAES detects 29.85%–320.08% more unique IDIs at 1.02×–28.04× higher throughput than the evaluated baselines, with 22.32×–43.32× lower redundancy. In multi-attribute testing, RAES detects 1.49×–7.53× more IDIs with 6.42×–231.50× lower redundancy. Retraining with RAES generated IDIs reduces discrimination from 0.045 to 0.016 while maintaining model accuracy. Conclusion: Controlling redundancy during test generation improves IDI detection and throughput for the tabular classifiers evaluated in this study.
- Open Access
A 25-Year Scoping Review of Integrated Theoretical Frameworks in Digital IT-Based Behavior Change Tools
Digital behavior change tools increasingly integrate constructs from multiple theoretical traditions, yet there is limited cross-domain understanding of how these frameworks are combined and which outcomes they are used to explain. This scoping review therefore aimed to map integrated theoretical frameworks used in digital IT-based behavior change tools from 1999 to 2025, focusing on theory combinations, outcome types, intervention contexts, and research gaps. Following PRISMA-ScR and a registered protocol (PROSPERO CRD42022285741), searches across six databases identified 62 eligible studies. Twenty-nine theories and models were identified, with Self-Determination Theory (35%) and the Theory of Planned Behavior (29%) being the most prevalent. Behavior-change-oriented frameworks were more commonly associated with target behavioral outcomes, whereas technology-adoption-oriented frameworks primarily explained intention, acceptance, continuance, and technology use. Health and fitness interventions dominated the evidence base (44%), followed by online learning (23%) and mobile commerce (11%). Long-term follow-up and explicit theory-to-behavior-change-technique mapping remained limited. Overall, the review provides a 25-year cross-domain synthesis of theoretical integration in digital behavior change tools and highlights the need for clearer theory–intervention alignment, longer-term evaluation, and broader application across underrepresented digital contexts.
Open AccessA 25-Year Scoping Review of Integrated Theoretical Frameworks in Digital IT-Based Behavior Change Tools
Digital behavior change tools increasingly integrate constructs from multiple theoretical traditions, yet there is limited cross-domain understanding of how these frameworks are combined and which outcomes they are used to explain. This scoping review therefore aimed to map integrated theoretical frameworks used in digital IT-based behavior change tools from 1999 to 2025, focusing on theory combinations, outcome types, intervention contexts, and research gaps. Following PRISMA-ScR and a registered protocol (PROSPERO CRD42022285741), searches across six databases identified 62 eligible studies. Twenty-nine theories and models were identified, with Self-Determination Theory (35%) and the Theory of Planned Behavior (29%) being the most prevalent. Behavior-change-oriented frameworks were more commonly associated with target behavioral outcomes, whereas technology-adoption-oriented frameworks primarily explained intention, acceptance, continuance, and technology use. Health and fitness interventions dominated the evidence base (44%), followed by online learning (23%) and mobile commerce (11%). Long-term follow-up and explicit theory-to-behavior-change-technique mapping remained limited. Overall, the review provides a 25-year cross-domain synthesis of theoretical integration in digital behavior change tools and highlights the need for clearer theory–intervention alignment, longer-term evaluation, and broader application across underrepresented digital contexts.
- Open Access
- 19 August 2026
- Published by Springer Nature in Journal of King Saud University Computer and Information Sciences
- Vol. 38 (7) , 685
FIFT: Feature Importance-Guided Fairness Testing for machine learning software
- 19 August 2026
- Published by Springer Nature in Journal of King Saud University Computer and Information Sciences
- Vol. 38 (7) , 685
Open AccessFIFT: Feature Importance-Guided Fairness Testing for machine learning software
- 19 June 2026
- Published by Elsevier in Knowledge-Based Systems
- Vol. 349
Toward causally-inspired explanations of maintenance effort: Integrating SHAP and LLM-based heuristic reasoning
- 19 June 2026
- Published by Elsevier in Knowledge-Based Systems
- Vol. 349
Toward causally-inspired explanations of maintenance effort: Integrating SHAP and LLM-based heuristic reasoning
- Open Access
- 9 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 37649-37667
Lossless Compression and Data Transformation Techniques on QR Code Binary Bit Stream
This paper presents a dataset-level evaluation of six lossless compression and data transformation techniques applied to visual-cryptographic (VC) shares derived from QR codes. We processed 40,000 QR samples, comprising 10,000 QR images (Versions 1-4, 2,500 per version), 10,000 QR images (Versions 1-10, 1,000 per version across ten application domains), and 20,000 augmented QR images (with noise, rotation, shear, cropping, and brightness variations). Each QR image is converted to VC share (share1), flattened to a bitstream, and evaluated under traditional compression techniques such as Run Length Encoding (RLE), Huffman Coding, Lempel Ziv-Welch (LZW), and data transformation techniques such as Binary-to-Integer, Base64 Encoding, and (BWT + MTF + Huffman Coding) Burrows Wheeler Transform (BWT), Move-To-Front (MTF), and Huffman Coding as a combined pipeline. Our experiments report Shannon entropy, compressed character count, compressed character count percentage, compression time, decompression time, memory usage, peak memory, lossless fidelity, metadata, payload size, storage size, and compression ratio. Empirical results show near-maximal entropy in QR-derived VC data (∼0.99), providing constraints on compression performance for traditional algorithms. Base64 consistently yields the best compression performance across both clean and augmented datasets, with an average compression rate of 499%. This work contributes a reproducible pipeline, a generalized dataset, and a benchmark reference for compression research on a highly randomized binary dataset.
- 9 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 37649-37667
Open AccessLossless Compression and Data Transformation Techniques on QR Code Binary Bit Stream
This paper presents a dataset-level evaluation of six lossless compression and data transformation techniques applied to visual-cryptographic (VC) shares derived from QR codes. We processed 40,000 QR samples, comprising 10,000 QR images (Versions 1-4, 2,500 per version), 10,000 QR images (Versions 1-10, 1,000 per version across ten application domains), and 20,000 augmented QR images (with noise, rotation, shear, cropping, and brightness variations). Each QR image is converted to VC share (share1), flattened to a bitstream, and evaluated under traditional compression techniques such as Run Length Encoding (RLE), Huffman Coding, Lempel Ziv-Welch (LZW), and data transformation techniques such as Binary-to-Integer, Base64 Encoding, and (BWT + MTF + Huffman Coding) Burrows Wheeler Transform (BWT), Move-To-Front (MTF), and Huffman Coding as a combined pipeline. Our experiments report Shannon entropy, compressed character count, compressed character count percentage, compression time, decompression time, memory usage, peak memory, lossless fidelity, metadata, payload size, storage size, and compression ratio. Empirical results show near-maximal entropy in QR-derived VC data (∼0.99), providing constraints on compression performance for traditional algorithms. Base64 consistently yields the best compression performance across both clean and augmented datasets, with an average compression rate of 499%. This work contributes a reproducible pipeline, a generalized dataset, and a benchmark reference for compression research on a highly randomized binary dataset.
- 8 November 2025
- Published by Springer Nature
- p. 269-280
How Mature is m-Learning in an Educational Institution?
- 8 November 2025
- Published by Springer Nature
- p. 269-280
How Mature is m-Learning in an Educational Institution?
- Open Access
Empirical Analysis of Data Sampling-Based Decision Forest Classifiers for Software Defect Prediction
The strategic significance of software testing in ensuring the success of software development projects is paramount. Comprehensive testing, conducted early and consistently across the development lifecycle, is vital for mitigating defects, especially given the constraints on time, budget, and other resources often faced by development teams. Software defect prediction (SDP) serves as a proactive approach to identifying software components that are most likely to be defective. By predicting these high-risk modules, teams can prioritize thorough testing and inspection, thereby preventing defects from escalating to later stages where resolution becomes more resource intensive. SDP models must be continuously refined to improve predictive accuracy and performance. This involves integrating clean and preprocessed datasets, leveraging advanced machine learning (ML) methods, and optimizing key metrics. Statistical-based and traditional ML approaches have been widely explored for SDP. However, statistical-based models often struggle with scalability and robustness, while conventional ML models face challenges with imbalanced datasets, limiting their prediction efficacy. In this study, innovative decision forest (DF) models were developed to address these limitations. Specifically, this study evaluates the cost-sensitive forest (CS-Forest), forest penalizing attributes (FPA), and functional trees (FT) as DF models. These models were further enhanced using homogeneous ensemble techniques, such as bagging and boosting techniques. The experimental analysis on benchmark SDP datasets demonstrates that the proposed DF models effectively handle class imbalance, accurately distinguishing between defective and non-defective modules. Compared to baseline and state-of-the-art ML and deep learning (DL) methods, the suggested DF models exhibit superior prediction performance and offer scalable solutions for SDP. Consequently, the application of DF-based models is recommended for advancing defect prediction in software engineering and similar ML domains.
Open AccessEmpirical Analysis of Data Sampling-Based Decision Forest Classifiers for Software Defect Prediction
The strategic significance of software testing in ensuring the success of software development projects is paramount. Comprehensive testing, conducted early and consistently across the development lifecycle, is vital for mitigating defects, especially given the constraints on time, budget, and other resources often faced by development teams. Software defect prediction (SDP) serves as a proactive approach to identifying software components that are most likely to be defective. By predicting these high-risk modules, teams can prioritize thorough testing and inspection, thereby preventing defects from escalating to later stages where resolution becomes more resource intensive. SDP models must be continuously refined to improve predictive accuracy and performance. This involves integrating clean and preprocessed datasets, leveraging advanced machine learning (ML) methods, and optimizing key metrics. Statistical-based and traditional ML approaches have been widely explored for SDP. However, statistical-based models often struggle with scalability and robustness, while conventional ML models face challenges with imbalanced datasets, limiting their prediction efficacy. In this study, innovative decision forest (DF) models were developed to address these limitations. Specifically, this study evaluates the cost-sensitive forest (CS-Forest), forest penalizing attributes (FPA), and functional trees (FT) as DF models. These models were further enhanced using homogeneous ensemble techniques, such as bagging and boosting techniques. The experimental analysis on benchmark SDP datasets demonstrates that the proposed DF models effectively handle class imbalance, accurately distinguishing between defective and non-defective modules. Compared to baseline and state-of-the-art ML and deep learning (DL) methods, the suggested DF models exhibit superior prediction performance and offer scalable solutions for SDP. Consequently, the application of DF-based models is recommended for advancing defect prediction in software engineering and similar ML domains.
- Open Access
- 26 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 46070-46090
Exploring the Integration of Generative AI Tools in Software Testing Education: A Case Study on ChatGPT and Copilot for Preparatory Testing Artifacts in Postgraduate Learning
Software testing education is important for building qualified testing professionals. To ensure that software testing graduates are ready for real-world challenges, it is necessary to integrate modern tools and technologies into the curriculum. With the emergence of Large Language Models (LLMs), their potential use in software engineering has become a focus, but their application in software testing education remains largely unexplored. This study, conducted in the Capstone Project course of a postgraduate software testing program, was carried out over two semesters with two distinct groups of students. A custom-built Travel Application limited to a web platform was used in the first semester. In the second semester, a new set of students worked with an open-source application, offering a larger-scale, multi-platform experience across web, desktop, and mobile platforms. Students initially created preparatory testing artifacts manually as a group deliverable. Following this, they were assigned an individual assignment to generate the same artifacts using LLM tools such as ChatGPT 3.5 in the first semester and Microsoft Copilot in the second. This process directly compared manually created artifacts and those generated using LLMs, leveraging AI for faster outputs. After completion, they responded to a set of assigned questions. The students’ responses were assessed using an integrated methodology, including quantitative and qualitative assessments, sentiment analysis to understand emotions, and a thematic approach to extract deeper insights. The findings revealed that while LLMs can assist and augment manual testing efforts, they cannot entirely replace the need for manual testing. By incorporating innovative technology into the curriculum, this study highlights how Generative AI can support active learning, connect theoretical concepts with practical applications, and align educational practices with industry needs.
- 26 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 46070-46090
Open AccessExploring the Integration of Generative AI Tools in Software Testing Education: A Case Study on ChatGPT and Copilot for Preparatory Testing Artifacts in Postgraduate Learning
Software testing education is important for building qualified testing professionals. To ensure that software testing graduates are ready for real-world challenges, it is necessary to integrate modern tools and technologies into the curriculum. With the emergence of Large Language Models (LLMs), their potential use in software engineering has become a focus, but their application in software testing education remains largely unexplored. This study, conducted in the Capstone Project course of a postgraduate software testing program, was carried out over two semesters with two distinct groups of students. A custom-built Travel Application limited to a web platform was used in the first semester. In the second semester, a new set of students worked with an open-source application, offering a larger-scale, multi-platform experience across web, desktop, and mobile platforms. Students initially created preparatory testing artifacts manually as a group deliverable. Following this, they were assigned an individual assignment to generate the same artifacts using LLM tools such as ChatGPT 3.5 in the first semester and Microsoft Copilot in the second. This process directly compared manually created artifacts and those generated using LLMs, leveraging AI for faster outputs. After completion, they responded to a set of assigned questions. The students’ responses were assessed using an integrated methodology, including quantitative and qualitative assessments, sentiment analysis to understand emotions, and a thematic approach to extract deeper insights. The findings revealed that while LLMs can assist and augment manual testing efforts, they cannot entirely replace the need for manual testing. By incorporating innovative technology into the curriculum, this study highlights how Generative AI can support active learning, connect theoretical concepts with practical applications, and align educational practices with industry needs.