Lívia Almada Cruz
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Lívia Almada Cruz has a total of 24 co-authors. Here are the top co-authors by number of shared publications:
- Ticiana Linhares Coelho da Silva
Federal University of Ceará (UFC)
4 shared publications - José Antônio Fernandes de Macêdo
Federal University of Ceará
3 shared publications - Régis Pires Magalhães
Universidade Federal do Ceará
3 shared publications - Anselmo Ramalho Pitombeira-Neto
Department of Industrial Engineering, Federal University of Cear, Fortaleza, Ceará, Brazil
2 shared publications - Karine Zeitouni
DAVID Lab, UVSQ-Université Paris-Saclay, Versailles, France
2 shared publications Daniel J.R. Vasconcelos
UFC, Insight Data Sci Lab, Sao Paulo, Brazil
1 shared publication
Latest Publications
- Open Access
- 1 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. PP (1530437X) , 1
Human Activity Recognition via a Categorical Kalman Filter Using Inertial Sensors
Human Activity Recognition (HAR) has gained significant attention due to its applications in healthcare, sports, and surveillance systems. While many existing models demonstrate good performance, they often lack interpretability and struggle with nonstationary data distributions common in sensor-based applications. This paper proposes a novel Categorical Kalman Filter (CatKF) for classifying human activities using inertial measurement unit (IMU) data. The CatKF extends the traditional Kalman filter framework to handle classification tasks by modeling class probabilities through a softmax function with time-varying parameters that evolve according to dynamic linear models. The approach enables online processing, adapts to distributional shifts, and maintains computational efficiency through sequential parameter updates. We evaluate the proposed method on three standard HAR datasets (PAMAP2, DSA, and USC-HAD) using leave-one-subject-out cross-validation. Experimental results demonstrate that the CatKF with learned hyperparameters achieves competitive performance compared to classical machine learning methods, matching multilayer perceptrons while being outperformed only by ensemble methods like Random Forest and XGBoost. A complementary online classification experiment further demonstrates fast adaptation and robustness under nonstationary data streams. We also analyze the temporal evolution of the model’s predictive parameters, highlighting how it adapts to new information and how its uncertainty changes over time. The Random-Walk Categorical Kalman Filter (RWCatKF) variant requires only three hyperparameters and operates without matrix inversions, making it particularly suitable for resource-constrained wearable devices. The proposed approach offers a transparent, interpretable alternative for HAR applications where model explainability is essential.
- 1 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. PP (1530437X) , 1
Open AccessHuman Activity Recognition via a Categorical Kalman Filter Using Inertial Sensors
Human Activity Recognition (HAR) has gained significant attention due to its applications in healthcare, sports, and surveillance systems. While many existing models demonstrate good performance, they often lack interpretability and struggle with nonstationary data distributions common in sensor-based applications. This paper proposes a novel Categorical Kalman Filter (CatKF) for classifying human activities using inertial measurement unit (IMU) data. The CatKF extends the traditional Kalman filter framework to handle classification tasks by modeling class probabilities through a softmax function with time-varying parameters that evolve according to dynamic linear models. The approach enables online processing, adapts to distributional shifts, and maintains computational efficiency through sequential parameter updates. We evaluate the proposed method on three standard HAR datasets (PAMAP2, DSA, and USC-HAD) using leave-one-subject-out cross-validation. Experimental results demonstrate that the CatKF with learned hyperparameters achieves competitive performance compared to classical machine learning methods, matching multilayer perceptrons while being outperformed only by ensemble methods like Random Forest and XGBoost. A complementary online classification experiment further demonstrates fast adaptation and robustness under nonstationary data streams. We also analyze the temporal evolution of the model’s predictive parameters, highlighting how it adapts to new information and how its uncertainty changes over time. The Random-Walk Categorical Kalman Filter (RWCatKF) variant requires only three hyperparameters and operates without matrix inversions, making it particularly suitable for resource-constrained wearable devices. The proposed approach offers a transparent, interpretable alternative for HAR applications where model explainability is essential.
- 30 January 2026
- Published by Springer Nature
- p. 455-469
RS4: Restricted Search Space for Snippet Selection
- 30 January 2026
- Published by Springer Nature
- p. 455-469
RS4: Restricted Search Space for Snippet Selection
- Open Access
- 13 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 30316-30333
An Ensemble Bayesian Dynamic Linear Model for Human Activity Recognition
Human activity recognition (HAR) has been gaining attention in recent years as a result of its many applications in health, sports, entertainment, and surveillance. Due to the current ubiquity of inertial measurement units (IMUs) in consumer electronics, including accelerometers and gyroscopes, HAR applications have increasingly used signals produced by these sensors. However, HAR from IMU data is challenging, since time-series data generated from human activity are typically multivariate, non-stationary, and noisy. In this work, we investigate the application of Bayesian dynamic linear models (BDLMs) to the online classification of time-series data of human activity acquired from IMUs. BDLMs are promising in HAR from IMU signals because they seamlessly handle temporal dependencies and uncertainty inherent in sensor data. Unlike static classifiers, BDLMs account for the sequential nature of IMU signals, enabling more accurate tracking of transitions between activities. In particular, we propose a method based on an ensemble BDLM for online HAR that is fully transparent and requires little preprocessing of data. We test the proposed method in two tasks, activity classification and intensity classification, and use two real datasets with diverse activities and subjects. The experimental results indicate that the proposed ensemble BDLM is competitive with consolidated benchmark methods and can be an effective method in real applications of online HAR.
- 13 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 30316-30333
Open AccessAn Ensemble Bayesian Dynamic Linear Model for Human Activity Recognition
Human activity recognition (HAR) has been gaining attention in recent years as a result of its many applications in health, sports, entertainment, and surveillance. Due to the current ubiquity of inertial measurement units (IMUs) in consumer electronics, including accelerometers and gyroscopes, HAR applications have increasingly used signals produced by these sensors. However, HAR from IMU data is challenging, since time-series data generated from human activity are typically multivariate, non-stationary, and noisy. In this work, we investigate the application of Bayesian dynamic linear models (BDLMs) to the online classification of time-series data of human activity acquired from IMUs. BDLMs are promising in HAR from IMU signals because they seamlessly handle temporal dependencies and uncertainty inherent in sensor data. Unlike static classifiers, BDLMs account for the sequential nature of IMU signals, enabling more accurate tracking of transitions between activities. In particular, we propose a method based on an ensemble BDLM for online HAR that is fully transparent and requires little preprocessing of data. We test the proposed method in two tasks, activity classification and intensity classification, and use two real datasets with diverse activities and subjects. The experimental results indicate that the proposed ensemble BDLM is competitive with consolidated benchmark methods and can be an effective method in real applications of online HAR.
- Open Access
- 17 January 2025
- Published by Sociedade Brasileira de Computação in Journal of the Brazilian Computer Society
- Vol. 31 (1) , 203-218
Machine Learning methods and models to predict food insecurity levels for families in Ceará, Brazil, based on employment, housing and other social indicators
Many nations still struggle to provide their populations with access to food and balanced nutrition. The Food and Agriculture Organization of the United Nations (FAO) included Brazil in its 2022 Hunger Map, highlighting that 61 million Brazilians face difficulties in feeding themselves. Despite the presence of various food security alert and monitoring systems in food-insecure countries, the data and methodologies they rely on capture only a fraction of the issue’s complexity, underscoring the need for further research to fully comprehend this multifaceted problem. In response, the Secretary for Social Protection of Ceará (SPS - Secretaria de Proteção Social), located in Brazil’s northeast, conducted a survey to collect data on the social and economic characteristics of extremely vulnerable families. This dataset, analyzed in our study, represents a concentrated effort by the government of Ceará to evaluate the needs of low-income households, particularly those with children who lack access to essential services. We used the Brazilian Food Insecurity Scale, a tool validated by the Brazilian Ministry, to measure food insecurity levels based on families’ responses, assigning scores to their answers. This paper presents a machine learning model that examines the collected data to identify which factors related to Food Access, Employment and Income, Housing, and Public Services can predict levels of food insecurity. Our best model demonstrates an accuracy of approximately 0.75, an F1-score of 0.80, and can distinguish between severe and non-severe food insecurity levels. We suggest that our model could be applied to other datasets lacking nutrition-specific questions to gauge a family’s food insecurity level. Additionally, our research sheds light on the key factors influencing food insecurity levels in Brazil, notably income and housing conditions, providing valuable insights for addressing this issue.
- 17 January 2025
- Published by Sociedade Brasileira de Computação in Journal of the Brazilian Computer Society
- Vol. 31 (1) , 203-218
Open AccessMachine Learning methods and models to predict food insecurity levels for families in Ceará, Brazil, based on employment, housing and other social indicators
Many nations still struggle to provide their populations with access to food and balanced nutrition. The Food and Agriculture Organization of the United Nations (FAO) included Brazil in its 2022 Hunger Map, highlighting that 61 million Brazilians face difficulties in feeding themselves. Despite the presence of various food security alert and monitoring systems in food-insecure countries, the data and methodologies they rely on capture only a fraction of the issue’s complexity, underscoring the need for further research to fully comprehend this multifaceted problem. In response, the Secretary for Social Protection of Ceará (SPS - Secretaria de Proteção Social), located in Brazil’s northeast, conducted a survey to collect data on the social and economic characteristics of extremely vulnerable families. This dataset, analyzed in our study, represents a concentrated effort by the government of Ceará to evaluate the needs of low-income households, particularly those with children who lack access to essential services. We used the Brazilian Food Insecurity Scale, a tool validated by the Brazilian Ministry, to measure food insecurity levels based on families’ responses, assigning scores to their answers. This paper presents a machine learning model that examines the collected data to identify which factors related to Food Access, Employment and Income, Housing, and Public Services can predict levels of food insecurity. Our best model demonstrates an accuracy of approximately 0.75, an F1-score of 0.80, and can distinguish between severe and non-severe food insecurity levels. We suggest that our model could be applied to other datasets lacking nutrition-specific questions to gauge a family’s food insecurity level. Additionally, our research sheds light on the key factors influencing food insecurity levels in Brazil, notably income and housing conditions, providing valuable insights for addressing this issue.
- 30 April 2023
- Published by Elsevier in Decision Support Systems
- Vol. 172
An analytical citizen relation management system (CZRM) for social vulnerability mapping and policy recommendation in Brazil
- 30 April 2023
- Published by Elsevier in Decision Support Systems
- Vol. 172
An analytical citizen relation management system (CZRM) for social vulnerability mapping and policy recommendation in Brazil
- Open Access
Modeling Trajectories Obtained from External Sensors for Location Prediction via NLP Approaches
Representation learning seeks to extract useful and low-dimensional attributes from complex and high-dimensional data. Natural language processing (NLP) was used to investigate the representation learning models to extract words’ feature vectors using their sequential order in the text via word embeddings and language models that maintain their semantic meaning. Inspired by NLP, in this paper, we tackle the representation learning problem for trajectories, using NLP methods to encode external sensors positioned in the road network and generate the features’ space to predict the next vehicle movement. We evaluate the vector representations of on-road sensors and trajectories using extrinsic and intrinsic strategies. Our results have shown the potential of natural language models to describe the space of features on trajectory applications as the next location prediction.
Open AccessModeling Trajectories Obtained from External Sensors for Location Prediction via NLP Approaches
Representation learning seeks to extract useful and low-dimensional attributes from complex and high-dimensional data. Natural language processing (NLP) was used to investigate the representation learning models to extract words’ feature vectors using their sequential order in the text via word embeddings and language models that maintain their semantic meaning. Inspired by NLP, in this paper, we tackle the representation learning problem for trajectories, using NLP methods to encode external sensors positioned in the road network and generate the features’ space to predict the next vehicle movement. We evaluate the vector representations of on-road sensors and trajectories using extrinsic and intrinsic strategies. Our results have shown the potential of natural language models to describe the space of features on trajectory applications as the next location prediction.
- 25 June 2020
- Published by Springer Nature in Distributed and Parallel Databases
- Vol. 39 (1) , 259-280
Location prediction: a deep spatiotemporal learning from external sensors data
- 25 June 2020
- Published by Springer Nature in Distributed and Parallel Databases
- Vol. 39 (1) , 259-280
Location prediction: a deep spatiotemporal learning from external sensors data