http://bibliotecatede.uninove.br/handle/tede/35 2026-06-06T08:14:48Z http://bibliotecatede.uninove.br/handle/tede/3986 Título: Novo índice clínico-laboratorial para predição de deterioração em adultos hospitalizados com modelo híbrido de aprendizagem profunda Autor: Oliveira, Reinaldo Ribeiro de Primeiro orientador: Dias, Cleber Gustavo Abstract: INTRODUCTION: Clinical deterioration in patients, which may lead to increased mortality risk, particularly in intensive care units, is associated with a dynamic process of physiological decline characterized by the progression of organ dysfunction. This process can be monitored over time through measurable indicators obtained from patients’ clinical records. Assessing clinical deterioration is a demanding task due to the complex and variable clinical behavior of each patient. The growing volume of clinical data within Electronic Health Records (EHRs) is characterized as longitudinal patient health information. These data enable the application of clinical deterioration protocols capable of observing and tracking physiological progression in hospitalized patients. The use of EHRs allows electronic records to generate insights and predictions of future clinical events. This study is justified by the need to develop a novel index grounded in hybrid artificial intelligence capable of processing heterogeneous temporal clinical data derived from electronic health records. OBJECTIVE: To develop and test an index composed of clinical and laboratory data for predicting deterioration in hospitalized adult patients using machine learning algorithms. METHODS: This study has an exploratory experimental technological design. A total of 1,100,000 patients were selected from two datasets originally extracted from the MIMIC-III database. The study was structured into three stages: (i) a scoping review, (ii) development of an adult patient deterioration index based on the most relevant clinical and laboratory time-series data treated statistically, and (iii) computational experiments using a hybrid approach combining convolutional neural networks (CNNs) and long short-term memory (LSTM) networks to predict the deterioration index at future time steps. Mathematical models were constructed to calculate the weights of variables composing the deterioration index. RESULTS: The hybrid neural network model was evaluated across 10 different scenarios with hyperparameter tuning (Seed, Past_H, Future_H, Val_Split, Epochs, and Batch_Size). The model demonstrated the ability to predict future events using retrospective time-series records to forecast 6-hour, 12-hour, 18-hour, and 24-hour horizons. Global performance metrics achieved were MAE ranging from 9.36 to 9.89 and MAPE from 24.19% to 25.96%. CONCLUSIONS: The hybrid model showed consistent performance under similar hyperparameters across different patient cohorts from MIMIC-III and MIMIC-IV. The best results were observed for 6-hour-ahead predictions. The study presents innovative potential and offers scientific contributions to the academic community, society, and healthcare delivery processes. The findings suggest that the proposed approach can anticipate clinical events and provide opportunities for timely interventions and clinical decision-making in hospital settings, thereby enhancing patient safety. Instituição: Universidade Nove de Julho Tipo do documento: Tese 2026-03-10T00:00:00Z http://bibliotecatede.uninove.br/handle/tede/3970 Título: Abordagem de busca local ativa em soluções não atrasadas para o problema de job shop scheduling baseado no algoritmo genético de chaves aleatórias Autor: Betoni, David Rios Primeiro orientador: Pereira, Fabio Henrique Abstract: In this work, a computational approach is presented for solving production task se-quencing problems, known as Job Shop Scheduling. It is an optimization problem that seeks to define the ideal sequence of operations for different tasks (jobs) across a variety of machines, with the objective of minimizing the total production time (makespan). In current literature, the use of metaheuristics has been widely adopted to solve this type of problem, especially Genetic Algorithms, as they enable the achievement of high-quality solutions. In general, however, metaheuristics require the joint use of solution refinement techniques. Among these techniques, Local Search methods stand out, aiming to improve the results obtained by the metaheuristics. This work fits into this context by proposing a two-stage hybrid approach. Initially, a variation of the Genetic Algorithm, known as the Random-Key Genetic Algorithm (RKGA), is used to generate a high-quality initial solu¬tion. Then, after the conclusion of the evolutionary process—that is, without incorporating local search mechanisms during the execution of the genetic algorithm—an independent refinement stage based on Local Search is applied. This second stage acts on the solution generated by the RKGA, exploring the reduced space of non-delay solutions, with the ob¬jective of improving the total production time. The results, obtained from standardized benchmark problem sets widely used in the literature, show that suboptimal non-delay solutions are frequently improved by the proposed refinement method, achieving results superior to those of the conventional genetic algorithm implementation. It is concluded, therefore, that the obtained results validate the proposed approach, demonstrating its effectiveness in achieving more efficient solutions. Instituição: Universidade Nove de Julho Tipo do documento: Dissertação 2026-03-02T00:00:00Z http://bibliotecatede.uninove.br/handle/tede/3920 Título: Inteligência artificial com processamento de linguagem natural em people analytics para predição de clima organizacional Autor: Vasconcellos, Sabrina da Silva Primeiro orientador: Belan, Peterson Adriano Abstract: Context: The growing adoption of artificial intelligence (AI) technologies in organizations has driven significant advances in how human behavior in the workplace is analyzed. Although areas such as Human Resources (HR) collect large volumes of natural language data such as exit interviews, feedback, climate surveys, and eNPS (Employee Net Promoter Score) these data are still not fully leveraged in people analytics. Objective: To apply natural language processing (NLP) techniques to develop predictive methods for organizational climate, using exit interviews as the textual basis. Method: To this end, a supervised sentiment analysis model based on Random Forest was conducted, with and without the use of synthetic data. The direct scale dimensions (1–5), (0–10) and sentiment labels (detractor, neutral, and promoter) were manually defined by three HR specialists (E1, E2, and E3) and consolidated through a consensus model (E4). One of the specialists was from the studied organization, while the others came from different organizations, broadening the market perspective. Indirect eNPS dimensions were also considered, manually defined by specialist E2, including leadership, career, communication, diversity, health and well-being, retention, teams, training, innovation, and engagement, and associated with the questions and responses from the interviews. Model performance was evaluated using metrics such as accuracy, precision, recall, and F1-score. Additionally, the study incorporated sentiment analyses grouped by indirect dimensions, applied statistics, visualizations, and temporal analyses, enabling observation of organizational climate evolution over time. Results: The final Random Forest model, trained exclusively on real data, achieved the best performance, reaching an accuracy of 75%. Conclusion: This result demonstrates the potential of the proposed approach to support data-driven strategic decisions, guide more targeted interventions, and contribute to more precise and human-centered people management in organizations. Instituição: Universidade Nove de Julho Tipo do documento: Dissertação 2025-12-16T00:00:00Z http://bibliotecatede.uninove.br/handle/tede/3919 Título: Detecção e mensuração de fissuras em edificações combinando o uso de drones com visão computacional e aprendizagem de máquina Autor: Paz, Gabriel Rosa Primeiro orientador: Araújo, Sidnei Alves de Abstract: The inspection of structures and buildings is an essential activity to ensure safety, performance, and durability in civil construction. However, traditional methods based on visual inspection present significant limitations, such as difficulty accessing elevated regions, exposure of professionals to risks, and dependence on the inspector’s interpretation. In this context, technologies based on drones, computer vision, and machine learning emerge as alternatives to automate the process and increase the reliability of the analysis. This work proposes a computational method for the automatic detection and measurement of cracks in buildings using images captured by drones. The method consists of two stages: the first involves automatic crack detection using a model of convolutional neural network, trained on images of cracks in civil works; the second stage performs the measurement of the detected crack (main length and predominant orientation), using techniques such as histogram of oriented gradients (HOG), skeletonization, and distance transform, as well as the conversion of pixel-based measurements into metric units. The results obtained in the detection stage (precision of 81.1%, recall of 69.7%, and mAP@50 of 71.1%) and in the measurement stage (average error of 2.24%), together with experiments using images from real building and civil construction environments and the simulation of a continuous monitoring scenario during drone flight, demonstrate the feasibility of the proposed method. Although opportunities for improvement were identified, such as expanding the training dataset and refining the calibration of the geometric measurement procedure, the present work advances the development of an integrated computational method that combines modern AI techniques, computer vision, and the use of drones to support inspections of buildings and civil engineering structures. Instituição: Universidade Nove de Julho Tipo do documento: Dissertação 2025-12-16T00:00:00Z