PPG - Pesquisa Operacional
URI Permanente para esta coleção
Navegar
Submissões Recentes
- ItemAcesso aberto (Open Access)Planning and deployment of wireless networks: a data-driven machine learning and optimization framework based on urban mesh and 5G networks(Universidade Federal de São Paulo, 2024-10-31) Jeske, Marlon [UNIFESP]; Nascimento, Mariá Cristina Vasconcelos; Aloise, Daniel; http://lattes.cnpq.br/5093210888872414; http://lattes.cnpq.br/1010810293243435; http://lattes.cnpq.br/4181573616012641Wireless networks play a fundamental role in the modern world, providing essential infrastructure for applications ranging from industrial automation to smart city development. Different wireless network architectures, such as mobile networks, wireless sensor networks, and wireless mesh networks, are used for different purposes and address specific requirements. Despite their distinct characteristics, these networks share common challenges in planning and deployment, particularly in complex urban environments where factors such as signal propagation, connectivity, and energy consumption must be carefully managed. This thesis addresses these challenges by proposing data-driven approaches based on machine learning and optimization techniques. It aims to fill critical gaps in the literature, particularly regarding accurate signal strength prediction and the optimal placement of relay devices. To achieve these objectives, three interrelated studies are presented in this thesis. In the first study, a machine learn
- ItemAcesso aberto (Open Access)Resilience in humanitarian operations: multi-methodological approach witch the FRAM and AHP-Gaussian methods(Universidade Federal de São Paulo, 2024-10-18) Pereira, Marcus Vinicius Gomes [UNIFESP]; Neto, Luiz Leduino de Salles [UNIFESP]; Santos, Marcos do; http://lattes.cnpq.br/5534398558592175; http://lattes.cnpq.br/3728820959678712; https://lattes.cnpq.br/0182704220595324This thesis investigates resilience in humanitarian operations, specifically in the context of Acolhida Operation (Brazil's Humanitarian Reception Operation) in the face of the Venezuelan migration crisis. The overall aim of the research is to develop an integrative multi-methodology that combines the FRAM (Functional Resonance Analysis Method) and AHP-Gaussian (Analytic Hierarchy Process-Gaussian) methods to improve the resilience of complex socio-technical systems. The choice of the FRAM and AHP-Gaussian methods is justified by the need for a comprehensive and structured analysis of complex systems and the prioritization of intervention alternatives. The research analyses the results of the FRAM and uses them as input for the AHP-Gaussian, comparing them in order to prioritize improvement measures and strengthen the resilience of Acolhida Operation. The approach adopted aims to provide a holistic view of the processes and interactions within the system, allowing vulnerabilities and areas for improvement to be identified with greater precision. The results indicate that the integration of these methods provides a analysis, allowing for a deeper understanding of the challenges faced by the operation. The research identified critical points and prioritized opportunities for improvement in Acolhida Operation, helping to reduce the human suffering of refugees and migrants and promoting a more efficient and dignified service. The application of AHP-Gaussian allows decisions to be based on multiple and weighted criteria, ensuring that the proposed interventions are not only effective, but also efficient and sustainable in the long term. In addition, an innovative framework was developed that can be applied to any complex socio-technical system. This framework allows for a continuous and adaptive assessment of systems, promoting the implementation of resilient practices that can be adjusted as new information and circumstances arise. The framework created not only facilitates the identification of failures and points for improvement, but also proposes solutions based on systematic prioritization. The contribution of this research lies in the innovative application of a multi-methodological approach to the analysis and improvement of complex systems, such as those of humanitarian operations, providing a solid basis for informed and strategic decisions, as well as allowing for an increase in the resilience of Acolhida Operation. This work advances knowledge and practice in the management of humanitarian operations, promoting more effective and adaptive strategies to respond to constantly evolving demands. The methodology developed has the potential to be applied in a variety of contexts, providing an instrument for seeking greater resilience in complex socio-technical systems.
- ItemAcesso aberto (Open Access)Machine learning for healthcare: a data-centric approach(Universidade Federal de São Paulo, 2024-06-25) Valeriano, Maria Gabriela [UNIFESP]; Lorena, Ana Carolina; Kiffer, Carlos Roberto Veiga [UNIFESP]; http://lattes.cnpq.br/7021893874375037; http://lattes.cnpq.br/3451628262694747; http://lattes.cnpq.br/7462488231975857Machine learning models have the potential to revolutionize the healthcare sector by leveraging continuously collected data in health systems. Traditionally, these models are trained on large datasets, with performance improvements achieved through robust models and hyperparameter tuning. In this work, we propose a data-centric approach focusing on improving the data itself. Throughout this research, a set of health-related databases was created. These databases originate from four distinct sources, encompassing the prediction of severe cases of COVID-19 and dengue, as well as the authorization of specialized care in the public health system in Brazil. The datasets created cover seven predictive tasks, each with separate training and testing data. All problems were designed as binary classification tasks and adopted tabular data. The datasets were initially characterized in relation to their hardness profiles, using a specific hardness measure proposed in previous works. This measure considers the probability of an instance being misclassified by different machine learning algorithms. Our analysis considered seven classifiers with distinct biases: Gradient Boosting, Random Forest, Logistic Regression, Multilayer Perceptron, Support Vector Classifier (with linear and RBF kernels), and Bagging. The models were evaluated using a set of metrics, area under the ROC curve and per-class recall and precision, to provide a holistic consideration of model performance. We proposed a new approach to generate post-hoc explanations for machine learning models. In this approach, we identified instances where the models are most likely to fail, offering data-centric explanations for such failures. The patterns found explain the model errors, resulting in greater confidence in the predictions made. Additionally, we present a case study where instance hardness analysis was adopted to improve the design of a prediction problem in collaboration with the data specialist. Our work demonstrated that through this approach, it was possible to improve data quality and, ultimately, model performance. Finally, we propose a generalized approach to enhance model performance when access to data experts is not possible. A two-step strategy was adopted: first, cleaning the training data based on instance difficulty values, and then introducing a reject option when the models did not offer high-confidence predictions for test instances. The results show that it is possible to improve model performance at the cost of rejecting instances from the test set.
- ItemAcesso aberto (Open Access)Hybrid model for selecting investment assets using the TODIM-θ method and Modern Portfolio Theory(Universidade Federal de São Paulo, 2024-07-02) Puppo, Bruna Dutra [UNIFESP]; Sbruzzi, Elton Felipe; Rangel, Luís Alberto Duncan; Leles, Michel Carlos Rodrigues; http://lattes.cnpq.br/9098047398813476; http://lattes.cnpq.br/5512914843540140; http://lattes.cnpq.br/0026358605322965; http://lattes.cnpq.br/9143172111876212This study presents the development of a hybrid model for the selection and optimization of investment portfolios, taking into account different investor profiles. The model employs the TODIM-θ method, a multi-criteria decision tool based on Prospect Theory and Modern Portfolio Theory, for optimization. The hybrid model was tested with real data from the stocks that make up the S&P 500 index between 2018 and 2022. It proved to be effective in handling large volumes of data and considering multiple alternatives and criteria, which makes it especially suitable for the selection of investments. The hybrid model represents a significant advance in the integration of the concepts of behavioral finance and optimization. By skillfully combining elements from both domains, the model builds portfolios that not only align with investor expectations but also achieve optimal results by adjusting their intrinsic values. Furthermore, the model can work quickly and efficiently, presenting results in a few minutes, without requiring high computational capacity. This demonstrates its practicality and applicability in the real world of investments.
- ItemAcesso aberto (Open Access)Semantic description and internal validation of clusters for applications in categorical data sets(Universidade Federal de São Paulo, 2024-06-19) Aquino, Roberto Douglas Guimarães de [UNIFESP]; Curtis, Vitor Venceslau; Verri, Filipe Alves Neto; http://lattes.cnpq.br/0145582312635382; http://lattes.cnpq.br/1785341067396776; http://lattes.cnpq.br/2373005809061037In clustering problems whose objective is not based specifically on spatial proximity but rather on feature patterns, traditional cluster validation indices may not be appropriate. This work proposes a tool that performs the description of clusters and can be used as an internal validation index to suggest the most appropriate number of clusters for applications in categorical data sets. To evaluate our index, we also propose a categorical synthetic data generator specifically designed for this application. We tested synthetic and real data sets with different configurations to evaluate the performance of the proposed index in comparison with well-known indexes in the literature. Thus, we demonstrate that the index has great potential to describe clusters and discover the number of most suitable clusters. The synthetic data generator is capable of producing relevant data sets for the internal validation process.