Roberta Andrade de Araújo Fagundes
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Roberta Andrade de Araújo Fagundes has a total of 72 co-authors. Here are the top co-authors by number of shared publications:
- Wylliams Barbosa Santos
Universidade de Pernambuco
7 shared publications - Halcyon Davys Pereira De Carvalho
Department of Computer Engineering, University of Pernambuco, Recife, Brazil
4 shared publications - Rafaella Leandra Souza Do Nascimento
Centro de Informática, Universidade Federal de Pernambuco, Recife, Brazil
4 shared publications - Ana Melo
Universidade de Pernambuco
3 shared publications - Marilia Lima
Universidade de Pernambuco
3 shared publications - Alexandre Magno Andrade Maciel
Universidade de Pernambuco
2 shared publications
Latest Publications
- Open Access
- 25 August 2026
- Published by Revista de Engenharia e Pesquisa Aplicada in Revista de Engenharia e Pesquisa Aplicada
- Vol. 11 (3) , 83-91
Análise Estatística e Comparativa de Algoritmos de Machine Learning para Predição de Mortalidade em Unidades de Terapia Intensiva
A predição de mortalidade em unidades de terapia intensiva (UTI) é um desafio crítico para a medicina. Este trabalho propõe a aplicação de algoritmos de Machine Learning para estimar o risco de óbito hospitalar, utilizando dados da base MIMIC-III processados com uma estratégia de interpolação inspirada no sistema MedLens. Foram avaliados cinco modelos supervisionados, com destaque para técnicas de ensemble como LightGBM e XGBoost. Diferentemente de abordagens anteriores, implementou-se um protocolo de validação rigoroso com 25 execuções independentes, seguido de testes de hipótese. Os resultados comprovaram a superioridade estatística do LightGBM, que atingiu acurácia média de 92,99% e F1-Score de 0,927. A aplicação do teste de Wilcoxon Signed-Rank confirmou ($p < 0,05$) que o desempenho do LightGBM é significativamente superior aos demais competidores, validando sua robustez como ferramenta de suporte à decisão clínica e otimização de recursos hospitalares.
- 25 August 2026
- Published by Revista de Engenharia e Pesquisa Aplicada in Revista de Engenharia e Pesquisa Aplicada
- Vol. 11 (3) , 83-91
Open AccessAnálise Estatística e Comparativa de Algoritmos de Machine Learning para Predição de Mortalidade em Unidades de Terapia Intensiva
A predição de mortalidade em unidades de terapia intensiva (UTI) é um desafio crítico para a medicina. Este trabalho propõe a aplicação de algoritmos de Machine Learning para estimar o risco de óbito hospitalar, utilizando dados da base MIMIC-III processados com uma estratégia de interpolação inspirada no sistema MedLens. Foram avaliados cinco modelos supervisionados, com destaque para técnicas de ensemble como LightGBM e XGBoost. Diferentemente de abordagens anteriores, implementou-se um protocolo de validação rigoroso com 25 execuções independentes, seguido de testes de hipótese. Os resultados comprovaram a superioridade estatística do LightGBM, que atingiu acurácia média de 92,99% e F1-Score de 0,927. A aplicação do teste de Wilcoxon Signed-Rank confirmou ($p < 0,05$) que o desempenho do LightGBM é significativamente superior aos demais competidores, validando sua robustez como ferramenta de suporte à decisão clínica e otimização de recursos hospitalares.
- Open Access
- 25 August 2026
- Published by Revista de Engenharia e Pesquisa Aplicada in Revista de Engenharia e Pesquisa Aplicada
- Vol. 11 (3) , 55-64
Comparação de Métodos de Otimização de Hiperparâmetros Baseados em Meta-Heurísticas e Estratégias Convencionais para Modelos de Detecção de Fraudes
A otimização de hiperparâmetros é importante na detecção de fraudes em cartões de crédito porque é um problemararo, dinâmico e altamente sensível a erros, e pequenos ajustes nos modelos podem gerar grandes diferenças nos resultados. Com isso, a otmização é crucial porqueeleva o desempenho, a robustez e a capacidade de capturar padrões raros, evitando tanto perdas financeiras quanto alarmes falsos. Assim, este estudo avalia diferentes estratégias de otimização de hiperparâmetros aplicadas à detecção de fraudes em cartões de crédito, um problema marcado por forte desbalanceamento e elevada complexidade. Técnicas tradicionais são comparadas à otimização bayesiana e a meta-heurísticas bioinspiradas. Para isso, foi utilizado o conjunto de dados Credit Card Fraud Detection e modelos de machine learning. O desempenho foi mensurado por métricas adequadas a eventos raros, como F1-Score, Recall e AUC-ROC, e a significância estatística foi analisada pelo teste de Wilcoxon. A comparação entre métodos revelou diferençasrelevantes e padrões que motivam uma análise mais aprofundada apresentada ao longo do estudo.
- 25 August 2026
- Published by Revista de Engenharia e Pesquisa Aplicada in Revista de Engenharia e Pesquisa Aplicada
- Vol. 11 (3) , 55-64
Open AccessComparação de Métodos de Otimização de Hiperparâmetros Baseados em Meta-Heurísticas e Estratégias Convencionais para Modelos de Detecção de Fraudes
A otimização de hiperparâmetros é importante na detecção de fraudes em cartões de crédito porque é um problemararo, dinâmico e altamente sensível a erros, e pequenos ajustes nos modelos podem gerar grandes diferenças nos resultados. Com isso, a otmização é crucial porqueeleva o desempenho, a robustez e a capacidade de capturar padrões raros, evitando tanto perdas financeiras quanto alarmes falsos. Assim, este estudo avalia diferentes estratégias de otimização de hiperparâmetros aplicadas à detecção de fraudes em cartões de crédito, um problema marcado por forte desbalanceamento e elevada complexidade. Técnicas tradicionais são comparadas à otimização bayesiana e a meta-heurísticas bioinspiradas. Para isso, foi utilizado o conjunto de dados Credit Card Fraud Detection e modelos de machine learning. O desempenho foi mensurado por métricas adequadas a eventos raros, como F1-Score, Recall e AUC-ROC, e a significância estatística foi analisada pelo teste de Wilcoxon. A comparação entre métodos revelou diferençasrelevantes e padrões que motivam uma análise mais aprofundada apresentada ao longo do estudo.
- Open Access
- 25 August 2026
- Published by Revista de Engenharia e Pesquisa Aplicada in Revista de Engenharia e Pesquisa Aplicada
- Vol. 11 (3) , 65-73
Clusterização Hierárquica de Mesorregiões com Base de Dados do SINAN: Um Estudo sobre Tuberculose no Estado de Pernambuco
A tuberculose permanece como desafio relevante de saúde pública no Brasil, com Pernambuco entre os estados de maior incidência. Este estudo analisou padrões epidemiológicos da doença entre 2001 e 2024, aplicando clusterização hierárquica às mesorregiões com dados do SINAN. O método de Ward, validado por métricas estatísticas e complementado pela PCA, identificou quatro agrupamentos robustos. A Região Metropolitana do Recife destacou-se como epicentro persistente, enquanto demais mesorregiões migraram de baixa/média para incidência intermediária. Observou-se predominância masculina (61–70%) em todos os clusters, especialmente nos de maior carga, além de diferenças etárias relevantes. A clusterização hierárquica mostrou-se vantajosa por revelar agrupamentos emmúltiplos níveis, oferecendo interpretações mais ricas e subsidiando estratégias de vigilância adaptadas ao perfil epidemiológico regional.
- 25 August 2026
- Published by Revista de Engenharia e Pesquisa Aplicada in Revista de Engenharia e Pesquisa Aplicada
- Vol. 11 (3) , 65-73
Open AccessClusterização Hierárquica de Mesorregiões com Base de Dados do SINAN: Um Estudo sobre Tuberculose no Estado de Pernambuco
A tuberculose permanece como desafio relevante de saúde pública no Brasil, com Pernambuco entre os estados de maior incidência. Este estudo analisou padrões epidemiológicos da doença entre 2001 e 2024, aplicando clusterização hierárquica às mesorregiões com dados do SINAN. O método de Ward, validado por métricas estatísticas e complementado pela PCA, identificou quatro agrupamentos robustos. A Região Metropolitana do Recife destacou-se como epicentro persistente, enquanto demais mesorregiões migraram de baixa/média para incidência intermediária. Observou-se predominância masculina (61–70%) em todos os clusters, especialmente nos de maior carga, além de diferenças etárias relevantes. A clusterização hierárquica mostrou-se vantajosa por revelar agrupamentos emmúltiplos níveis, oferecendo interpretações mais ricas e subsidiando estratégias de vigilância adaptadas ao perfil epidemiológico regional.
- 15 January 2026
- Published by Springer Nature
- p. 127-139
Application of Machine Learning in the Detection and Prevention of Femicide: A Systematic Review
- 15 January 2026
- Published by Springer Nature
- p. 127-139
Application of Machine Learning in the Detection and Prevention of Femicide: A Systematic Review
- 1 December 2025
- Published by Associacao Brasileira de Inteligencia Computacional - ABRICOM
- p. 1-8
Ensembles for Regression: A Statistical Analysis and Recommendations
Ensemble techniques have proven effective in improving the performance of regression models. However, the decision between homogeneous and heterogeneous strategies is often made empirically, without considering the statistical properties of the data. This paper presents a framework that conducts a statistical analysis of datasets to recommend the most appropriate ensemble strategy. The framework leverages metrics such as coefficient of variation, skewness, kurtosis, correlation, and outlier presence to classify datasets into homogeneous or heterogeneous profiles. Ten real-world datasets with varying levels of complexity were evaluated. The results show that the proposed framework achieved the best performance in 80% of the cases, confirming that the statistical structure of the data should guide the selection of ensemble strategies. Wilcoxon statistical tests further validated the significance of the results. The proposed framework offers a systematic alternative to guide ensemble decisions in regression tasks, enhancing accuracy and reducing reliance on empirical trial-and-error procedures.
- 1 December 2025
- Published by Associacao Brasileira de Inteligencia Computacional - ABRICOM
- p. 1-8
Ensembles for Regression: A Statistical Analysis and Recommendations
Ensemble techniques have proven effective in improving the performance of regression models. However, the decision between homogeneous and heterogeneous strategies is often made empirically, without considering the statistical properties of the data. This paper presents a framework that conducts a statistical analysis of datasets to recommend the most appropriate ensemble strategy. The framework leverages metrics such as coefficient of variation, skewness, kurtosis, correlation, and outlier presence to classify datasets into homogeneous or heterogeneous profiles. Ten real-world datasets with varying levels of complexity were evaluated. The results show that the proposed framework achieved the best performance in 80% of the cases, confirming that the statistical structure of the data should guide the selection of ensemble strategies. Wilcoxon statistical tests further validated the significance of the results. The proposed framework offers a systematic alternative to guide ensemble decisions in regression tasks, enhancing accuracy and reducing reliance on empirical trial-and-error procedures.
- 1 December 2025
- Published by Associacao Brasileira de Inteligencia Computacional - ABRICOM
- p. 1-8
Quality Prediction of Resistance Spot Welding with Preheating in Imbalanced Dataset
Resistance Spot Welding (RSW) holds significant economic importance across various industries, particularly in the automotive sector, where it plays a crucial role in ensuring the safety and structural integrity of manufactured automobiles. However, when working with galvanized metal sheets, RSW faces unique challenges due to the presence of zinc oxide layer on the surface, which can reduce weld quality. Therefore, a commonly adopted solution is the preheating technique, which helps disrupt this oxide layer and improve weld formation. Despite its widespread use, preheating is often overlooked in predictive quality prediction models. In response, this research addresses that gap by investigating the ability of machine learning techniques to classify welding process data between ‘good’ or ‘defective’ welds with information from preheating phase. To further improve classification performance—especially in the presence of imbalanced datasets—this study proposes an ensemble-based approach combined with characteristics extraction from physical features of the dynamic resistance curve. An ensemble approach was explored by integrating various base classifiers within a Balanced Bagging Classifier, combined through a Voting Classifier to leverage their complementary strengths. The performance of the ensemble model was rigorously evaluated using cross-validation technique to ensure robustness and generalizability. The results revealed that the use of balanced bagging classifiers improved the results of classifying defective and good spot welding with respect to default bagging classifiers in the context of imbalaced data.
- 1 December 2025
- Published by Associacao Brasileira de Inteligencia Computacional - ABRICOM
- p. 1-8
Quality Prediction of Resistance Spot Welding with Preheating in Imbalanced Dataset
Resistance Spot Welding (RSW) holds significant economic importance across various industries, particularly in the automotive sector, where it plays a crucial role in ensuring the safety and structural integrity of manufactured automobiles. However, when working with galvanized metal sheets, RSW faces unique challenges due to the presence of zinc oxide layer on the surface, which can reduce weld quality. Therefore, a commonly adopted solution is the preheating technique, which helps disrupt this oxide layer and improve weld formation. Despite its widespread use, preheating is often overlooked in predictive quality prediction models. In response, this research addresses that gap by investigating the ability of machine learning techniques to classify welding process data between ‘good’ or ‘defective’ welds with information from preheating phase. To further improve classification performance—especially in the presence of imbalanced datasets—this study proposes an ensemble-based approach combined with characteristics extraction from physical features of the dynamic resistance curve. An ensemble approach was explored by integrating various base classifiers within a Balanced Bagging Classifier, combined through a Voting Classifier to leverage their complementary strengths. The performance of the ensemble model was rigorously evaluated using cross-validation technique to ensure robustness and generalizability. The results revealed that the use of balanced bagging classifiers improved the results of classifying defective and good spot welding with respect to default bagging classifiers in the context of imbalaced data.
- 1 June 2025
- Published by Elsevier in Expert Systems with Applications
- Vol. 290
Dynamic selection of ensemble-based regression models: Systematic literature review
- 1 June 2025
- Published by Elsevier in Expert Systems with Applications
- Vol. 290
Dynamic selection of ensemble-based regression models: Systematic literature review
- Open Access
- 19 May 2025
- Published by Sociedade Brasileira de Computação
- p. 182-190
Analysis of Ensembles Applied to Time Series Forecasting of Hydroelectric Dams
Context: Dams are essential for electricity generation in Brazil. Given the countrys vast territorial extent, it is crucial to optimize the use of its energy sources, particularly the hydrological potential of rivers. Problem: Flow forecasting in river basins is critical for the safety and efficiency of operations. Underestimation errors can increase the risk of dam overflow and failure, while overestimations compromise water storage, affecting supply and power generation during dry periods. Solution: This study proposes a dynamic selection method of machine learning models to achieve more accurate and stable flow forecasts. IS Theory: The research is based on Dynamic Capabilities Theory, emphasizing the ability for internal reconfiguration to address complex issues such as time series forecasting. Method: Data were initially normalized and temporally adjusted to ensure consistency. A diverse set of base models was selected based on their intrinsic characteristics, then integrated into a dynamic ensemble framework. Performance was evaluated using MAPE and RMSE metrics, allowing comparative analysis between the proposed dynamic ensemble and traditional deep learning models. Results: The dynamic selection approach showed significant improvements, achieving errors of 23.88 (MAPE) and 852.97 (RMSE), outperforming traditional and deep learning models. Contributions and Impact in the IS Field: This work demonstrates that dynamic model selection is a superior and promising approach compared to isolated model use, offering a significant advancement for flow forecasting in dam systems and contributing to more efficient water resource management in the Brazilian context.
- 19 May 2025
- Published by Sociedade Brasileira de Computação
- p. 182-190
Open AccessAnalysis of Ensembles Applied to Time Series Forecasting of Hydroelectric Dams
Context: Dams are essential for electricity generation in Brazil. Given the countrys vast territorial extent, it is crucial to optimize the use of its energy sources, particularly the hydrological potential of rivers. Problem: Flow forecasting in river basins is critical for the safety and efficiency of operations. Underestimation errors can increase the risk of dam overflow and failure, while overestimations compromise water storage, affecting supply and power generation during dry periods. Solution: This study proposes a dynamic selection method of machine learning models to achieve more accurate and stable flow forecasts. IS Theory: The research is based on Dynamic Capabilities Theory, emphasizing the ability for internal reconfiguration to address complex issues such as time series forecasting. Method: Data were initially normalized and temporally adjusted to ensure consistency. A diverse set of base models was selected based on their intrinsic characteristics, then integrated into a dynamic ensemble framework. Performance was evaluated using MAPE and RMSE metrics, allowing comparative analysis between the proposed dynamic ensemble and traditional deep learning models. Results: The dynamic selection approach showed significant improvements, achieving errors of 23.88 (MAPE) and 852.97 (RMSE), outperforming traditional and deep learning models. Contributions and Impact in the IS Field: This work demonstrates that dynamic model selection is a superior and promising approach compared to isolated model use, offering a significant advancement for flow forecasting in dam systems and contributing to more efficient water resource management in the Brazilian context.
- Open Access
- 19 May 2025
- Published by Sociedade Brasileira de Computação
- p. 359-368
Enhancing Alzheimer’s Disease Diagnosis: Insights from MLP and 1D CNN Models
Context: Alzheimer’s Disease (AD) is a complex neurodegenerative disorder that requires early diagnosis to improve patient outcomes. Recent advances in computational intelligence have sparked interest in leveraging machine learning to enhance diagnostic accuracy and efficiency. These innovations are crucial for transforming decision-making within Information Systems in clinical settings. Problem: Traditional methods like PET-scans and cerebrospinal fluid collection are highly accurate but costly and invasive, limiting accessibility. Developing data-driven, non-invasive solutions that retain diagnostic accuracy while handling complex biomedical data, such as plasma protein concentrations, remains a challenge. Solution: This study utilizes neural networks, specifically Multi-Layer Perceptron (MLP) and One-Dimensional Convolutional Neural Network (1D CNN). Preprocessing included Recursive Feature Elimination (RFE) for feature selection and Synthetic Minority Oversampling Technique (SMOTE) for data augmentation, addressing class imbalance. SI Theory: Grounded in Complexity Theory, the study examines how machine learning models can enhance data-driven medical systems by efficiently managing critical, highly sensitive datasets. Method: An experimental quantitative approach was used to evaluate binary and multiclass classifiers on a dataset with 120 protein features from 259 patients. Summary of Results: The MLP exhibited strong performance in specific subsets, achieving superior metrics in the binary classification after feature selection and data augmentation. Meanwhile, the 1D CNN excelled in multiclass classification, leveraging its convolutional layers to extract critical features from subtle protein variations, improving accuracy and robustness. Contributions and Impact on IS Field: This research enhances medical information systems by proposing machine learning models that can be integrated for accurate diagnostics, supporting clinical decision-making and advancing healthcare practices.
- 19 May 2025
- Published by Sociedade Brasileira de Computação
- p. 359-368
Open AccessEnhancing Alzheimer’s Disease Diagnosis: Insights from MLP and 1D CNN Models
Context: Alzheimer’s Disease (AD) is a complex neurodegenerative disorder that requires early diagnosis to improve patient outcomes. Recent advances in computational intelligence have sparked interest in leveraging machine learning to enhance diagnostic accuracy and efficiency. These innovations are crucial for transforming decision-making within Information Systems in clinical settings. Problem: Traditional methods like PET-scans and cerebrospinal fluid collection are highly accurate but costly and invasive, limiting accessibility. Developing data-driven, non-invasive solutions that retain diagnostic accuracy while handling complex biomedical data, such as plasma protein concentrations, remains a challenge. Solution: This study utilizes neural networks, specifically Multi-Layer Perceptron (MLP) and One-Dimensional Convolutional Neural Network (1D CNN). Preprocessing included Recursive Feature Elimination (RFE) for feature selection and Synthetic Minority Oversampling Technique (SMOTE) for data augmentation, addressing class imbalance. SI Theory: Grounded in Complexity Theory, the study examines how machine learning models can enhance data-driven medical systems by efficiently managing critical, highly sensitive datasets. Method: An experimental quantitative approach was used to evaluate binary and multiclass classifiers on a dataset with 120 protein features from 259 patients. Summary of Results: The MLP exhibited strong performance in specific subsets, achieving superior metrics in the binary classification after feature selection and data augmentation. Meanwhile, the 1D CNN excelled in multiclass classification, leveraging its convolutional layers to extract critical features from subtle protein variations, improving accuracy and robustness. Contributions and Impact on IS Field: This research enhances medical information systems by proposing machine learning models that can be integrated for accurate diagnostics, supporting clinical decision-making and advancing healthcare practices.
- Open Access
- 30 September 2024
- Published by Sociedade Brasileira de Computação
- p. 665-671
Understanding the Relationship of Conflict, Psychological safety, and Success in Software Development Projects
Software development involves complex socio-technical activities with common conflicts arising from human and non-human factors. However, the impact of these conflicts, particularly non-human ones, on project success still needs to be investigated. Psychological safety is a key to mitigating these conflicts. In this paper, we aim to understand how conflicts and psychological safety impact the success of software development projects—considering the categorization of conflicts into human-rooted (HRC) and non-human-rooted (NHRC) and examining organization size, team size, and psychological safety as moderating factors. To achieve our goal, we replicated an existing study in the area and included new hypotheses; we also adapted the surveys to Portuguese, used a structured survey to collect data, and applied structural equation modeling for the analysis. The results include the reception of 155 responses and show that human-rooted conflict strongly negatively impacts software project success, regardless of organization or team size. Non-human-rooted conflict positively affects project success only in corporations (> 100 employees) but negatively impacts small and medium-sized organizations (< 100 employees). While not directly correlated with project success, psychological safety appears as a crucial moderating factor.
- 30 September 2024
- Published by Sociedade Brasileira de Computação
- p. 665-671
Open AccessUnderstanding the Relationship of Conflict, Psychological safety, and Success in Software Development Projects
Software development involves complex socio-technical activities with common conflicts arising from human and non-human factors. However, the impact of these conflicts, particularly non-human ones, on project success still needs to be investigated. Psychological safety is a key to mitigating these conflicts. In this paper, we aim to understand how conflicts and psychological safety impact the success of software development projects—considering the categorization of conflicts into human-rooted (HRC) and non-human-rooted (NHRC) and examining organization size, team size, and psychological safety as moderating factors. To achieve our goal, we replicated an existing study in the area and included new hypotheses; we also adapted the surveys to Portuguese, used a structured survey to collect data, and applied structural equation modeling for the analysis. The results include the reception of 155 responses and show that human-rooted conflict strongly negatively impacts software project success, regardless of organization or team size. Non-human-rooted conflict positively affects project success only in corporations (> 100 employees) but negatively impacts small and medium-sized organizations (< 100 employees). While not directly correlated with project success, psychological safety appears as a crucial moderating factor.