Elaine Parros Machado de Sousa
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Elaine Parros Machado de Sousa has a total of 95 co-authors. Here are the top co-authors by number of shared publications:
Agma J. M. Traina
5 shared publications- André Moreira Souza
University of São Paulo (USP)
4 shared publications - Agma Juci Machado Traina
University of São Paulo (USP)
3 shared publications Caetano Traina
University of São Paulo
3 shared publicationsJ. Zullo
Center of Meteorological and Climate Researches Applied to Agriculture (Cepagri), University of Campinas (Unicamp), Campinas, Brazil
3 shared publicationsL. A. S. Romani
Embrapa Agricultural Informatics, Campinas, Brazil
3 shared publications
Latest Publications
- 29 September 2026
- Published by Springer Nature
- p. 138-147
SmartIVF: A Z-Order Inverted File for Filtered Approximate Nearest-Neighbor Search
- 29 September 2026
- Published by Springer Nature
- p. 138-147
SmartIVF: A Z-Order Inverted File for Filtered Approximate Nearest-Neighbor Search
- Open Access
- 29 September 2025
- Published by Sociedade Brasileira de Computação
- p. 182-188
Handling of missing values in wearable data streams
The increasing volume of data generated by interconnected devices has amplified the need for efficient stream mining methods, particularly in healthcare applications using wearable devices. These systems enable continuous health monitoring and support early interventions. However, missing values—common in streaming data—can lead to biased or invalid decisions, especially when their underlying causes are unknown. This PhD project proposes a preprocessing method to handle missing values in healthcare wearable data streams, addressing challenges such as data evolution, diverse missing mechanisms, and computational constraints. The goal is to improve data quality and the robustness of mining tasks in real-time health monitoring systems.
- 29 September 2025
- Published by Sociedade Brasileira de Computação
- p. 182-188
Open AccessHandling of missing values in wearable data streams
The increasing volume of data generated by interconnected devices has amplified the need for efficient stream mining methods, particularly in healthcare applications using wearable devices. These systems enable continuous health monitoring and support early interventions. However, missing values—common in streaming data—can lead to biased or invalid decisions, especially when their underlying causes are unknown. This PhD project proposes a preprocessing method to handle missing values in healthcare wearable data streams, addressing challenges such as data evolution, diverse missing mechanisms, and computational constraints. The goal is to improve data quality and the robustness of mining tasks in real-time health monitoring systems.
- Open Access
- 29 September 2025
- Published by Sociedade Brasileira de Computação
- p. 344-356
eTRUMiner: Mining Multivariate Temporal Rules from Heterogeneous and Incomplete Time Series
This paper introduces eTRUMiner, a novel algorithm for mining multivariate temporal association rules from heterogeneous time series datasets with missing observations. Our approach enables user-defined discretization, discovers frequent patterns, and outputs rules in short or detailed formats, thus enhancing interpretability. The results show that: (i) the choice of discretization method significantly influences rule relevance; (ii) eTRUMiner preserves the rules with high confidence in a dataset with up to 15% missing data; and (iii) the extracted rules can capture plausible causal dynamics. These findings demonstrate eTRUMiner’s robustness to incomplete data and its usefulness for exploratory analysis and forecasting in complex temporal domains.
- 29 September 2025
- Published by Sociedade Brasileira de Computação
- p. 344-356
Open AccesseTRUMiner: Mining Multivariate Temporal Rules from Heterogeneous and Incomplete Time Series
This paper introduces eTRUMiner, a novel algorithm for mining multivariate temporal association rules from heterogeneous time series datasets with missing observations. Our approach enables user-defined discretization, discovers frequent patterns, and outputs rules in short or detailed formats, thus enhancing interpretability. The results show that: (i) the choice of discretization method significantly influences rule relevance; (ii) eTRUMiner preserves the rules with high confidence in a dataset with up to 15% missing data; and (iii) the extracted rules can capture plausible causal dynamics. These findings demonstrate eTRUMiner’s robustness to incomplete data and its usefulness for exploratory analysis and forecasting in complex temporal domains.
- 17 April 2025
- Published by Springer Nature in Applied Intelligence
- Vol. 55 (7) , 1-12
Deep learning solutions for audio event detection in a swine barn using environmental audio and weak labels
- 17 April 2025
- Published by Springer Nature in Applied Intelligence
- Vol. 55 (7) , 1-12
Deep learning solutions for audio event detection in a swine barn using environmental audio and weak labels
- 29 March 2025
- Published by Elsevier in Expert Systems with Applications
- Vol. 278
PATSA-BIL: Pipeline for automated texture and structure analysis of borehole image logs
- 29 March 2025
- Published by Elsevier in Expert Systems with Applications
- Vol. 278
PATSA-BIL: Pipeline for automated texture and structure analysis of borehole image logs
- 2 November 2024
- Published by Springer Nature in Knowledge and Information Systems
- Vol. 67 (2) , 1455-1479
CETra: online cluster tracking for clustering of streaming data sources
- 2 November 2024
- Published by Springer Nature in Knowledge and Information Systems
- Vol. 67 (2) , 1455-1479
CETra: online cluster tracking for clustering of streaming data sources
- Open Access
- 14 October 2024
- Published by Sociedade Brasileira de Computação
- p. 165-171
Análise de similaridades e padrões estatísticos em séries temporais de casos de suicídio no Brasil
Suicide is a serious global public health problem and is among the leading causes of death in the world. In Brazil, suicide rates have seen alarming increases and continue to rise. Some public policies have been implemented to prevent the rise in suicide cases. Despite this, studies in the field indicate that such measures are not enough to cause a reduction in cases. The collection and analysis of suicide data are important for creating and reevaluating monitoring and prevention processes for suicide attempts. The goal of this master’s project is to contribute to the recognition of temporal patterns of suicide cases in Brazil from suicide deaths data published by DATASUS, using a time series-based approach. The main steps of this work are the collection of data and modeling of time series, provision of a pre-processed database containing daily suicide cases data from Brazil, exploratory analysis of the time series, similarity search between time series and application of clustering; mining of time series subsequence patterns, integration of multimodal data, and finally, the evaluation of the results obtained. It is expected to identify inherent temporal patterns in the time series to assist in understanding the behavior of suicide cases in the states and regions of Brazil.
- 14 October 2024
- Published by Sociedade Brasileira de Computação
- p. 165-171
Open AccessAnálise de similaridades e padrões estatísticos em séries temporais de casos de suicídio no Brasil
Suicide is a serious global public health problem and is among the leading causes of death in the world. In Brazil, suicide rates have seen alarming increases and continue to rise. Some public policies have been implemented to prevent the rise in suicide cases. Despite this, studies in the field indicate that such measures are not enough to cause a reduction in cases. The collection and analysis of suicide data are important for creating and reevaluating monitoring and prevention processes for suicide attempts. The goal of this master’s project is to contribute to the recognition of temporal patterns of suicide cases in Brazil from suicide deaths data published by DATASUS, using a time series-based approach. The main steps of this work are the collection of data and modeling of time series, provision of a pre-processed database containing daily suicide cases data from Brazil, exploratory analysis of the time series, similarity search between time series and application of clustering; mining of time series subsequence patterns, integration of multimodal data, and finally, the evaluation of the results obtained. It is expected to identify inherent temporal patterns in the time series to assist in understanding the behavior of suicide cases in the states and regions of Brazil.
- Open Access
- 14 October 2024
- Published by Sociedade Brasileira de Computação
- p. 12-22
BrSuicides: Conjunto de dados para análise de casos diários de suicídio no Brasil
O suicídio está entre as principais causas de morte no mundo e com taxas crescentes no Brasil. A OMS e profissionais de saúde defendem que a coleta e a análise dos dados de suicídio são importantes para implementação e avaliação de regulações e políticas públicas para auxiliar na prevenção do suicídio. O Brasil, no entanto, ainda carece de análises aprofundadas e de implementação de políticas públicas para prevenção de suicídio. Nesse contexto, este trabalho apresenta e disponibiliza o conjunto BrSuicides, composto por dados relacionados a casos de suicídio no Brasil e criado a partir da base de dados de óbitos fornecida pelo DATASUS. As tarefas de preprocessamento reduziram significativamente as inconsistências presentes nos dados coletados e resultou num conjunto preparado para análise sob diferentes perspectivas, de análise estatística exploratória a aprendizado de máquina em séries temporais.
- 14 October 2024
- Published by Sociedade Brasileira de Computação
- p. 12-22
Open AccessBrSuicides: Conjunto de dados para análise de casos diários de suicídio no Brasil
O suicídio está entre as principais causas de morte no mundo e com taxas crescentes no Brasil. A OMS e profissionais de saúde defendem que a coleta e a análise dos dados de suicídio são importantes para implementação e avaliação de regulações e políticas públicas para auxiliar na prevenção do suicídio. O Brasil, no entanto, ainda carece de análises aprofundadas e de implementação de políticas públicas para prevenção de suicídio. Nesse contexto, este trabalho apresenta e disponibiliza o conjunto BrSuicides, composto por dados relacionados a casos de suicídio no Brasil e criado a partir da base de dados de óbitos fornecida pelo DATASUS. As tarefas de preprocessamento reduziram significativamente as inconsistências presentes nos dados coletados e resultou num conjunto preparado para análise sob diferentes perspectivas, de análise estatística exploratória a aprendizado de máquina em séries temporais.
- Open Access
Quantifying With Only Positive Training Data
Quantification is the research field that studies methods for counting the number of data points that belong to each class in an unlabeled sample. Traditionally, researchers in this field assume the availability of labelled observations for all classes to induce a quantification model. However, we often face situations where the number of classes is large or even unknown, or we have reliable data for a single class. When inducing a multi-class quantifier is infeasible, we are often concerned with estimates for a specific class of interest. In this context, we have proposed a novel setting known as One-class Quantification (OCQ). In contrast, Positive and Unlabeled Learning (PUL), another branch of Machine Learning, has offered solutions to OCQ, despite quantification not being the focal point of PUL. This article closes the gap between PUL and OCQ and brings both areas together under a unified view. We compare our method, Passive Aggressive Threshold (PAT), against PUL methods and show that PAT generally is the fastest and most accurate algorithm. PAT induces quantification models that can be reused to quantify different samples of data. We additionally introduce Exhaustive TIcE (ExTIcE), an improved version of the PUL algorithm Tree Induction for c Estimation (TIcE). We show that ExTIcE quantifies more accurately than PAT and the other assessed algorithms in scenarios where several negative observations are identical to the positive ones.
Open AccessQuantifying With Only Positive Training Data
Quantification is the research field that studies methods for counting the number of data points that belong to each class in an unlabeled sample. Traditionally, researchers in this field assume the availability of labelled observations for all classes to induce a quantification model. However, we often face situations where the number of classes is large or even unknown, or we have reliable data for a single class. When inducing a multi-class quantifier is infeasible, we are often concerned with estimates for a specific class of interest. In this context, we have proposed a novel setting known as One-class Quantification (OCQ). In contrast, Positive and Unlabeled Learning (PUL), another branch of Machine Learning, has offered solutions to OCQ, despite quantification not being the focal point of PUL. This article closes the gap between PUL and OCQ and brings both areas together under a unified view. We compare our method, Passive Aggressive Threshold (PAT), against PUL methods and show that PAT generally is the fastest and most accurate algorithm. PAT induces quantification models that can be reused to quantify different samples of data. We additionally introduce Exhaustive TIcE (ExTIcE), an improved version of the PUL algorithm Tree Induction for c Estimation (TIcE). We show that ExTIcE quantifies more accurately than PAT and the other assessed algorithms in scenarios where several negative observations are identical to the positive ones.
- 29 August 2021
- Published by Wiley in Journal of Animal Breeding and Genetics
- Vol. 139 (1) , 100-112
Practical implications of using non‐relational databases to store large genomic data files and novel phenotypes
The objective of our study was to provide practical directions on the storage of genomic information and novel phenotypes (treated here as unstructured data) using a non‐relational database. The MongoDB technology was assessed for this purpose, enabling frequent data transactions involving numerous individuals under genetic evaluation. Our study investigated different genomic (Illumina Final Report, PLINK, 0125, FASTQ, and VCF formats) and phenotypic (including media files) information, using both real and simulated datasets. Advantages of our centralized database concept include the sublinear running time for queries after increasing the number of samples/markers exponentially, in addition to the comprehensive management of distinct data formats while searching for specific genomic regions. A comparison of our non‐relational and generic solution, with an existing relational approach (developed for tabular data types using 2 bits to store genotypes), showed reduced importing time to handle 50M SNPs (PLINK format) achieved by the relational schema. Our experimental results also reinforce that data conversion is a costly step required to manage genomic data into both relational and non‐relational database systems, and therefore, must be carefully treated for large applications.
- 29 August 2021
- Published by Wiley in Journal of Animal Breeding and Genetics
- Vol. 139 (1) , 100-112
Practical implications of using non‐relational databases to store large genomic data files and novel phenotypes
The objective of our study was to provide practical directions on the storage of genomic information and novel phenotypes (treated here as unstructured data) using a non‐relational database. The MongoDB technology was assessed for this purpose, enabling frequent data transactions involving numerous individuals under genetic evaluation. Our study investigated different genomic (Illumina Final Report, PLINK, 0125, FASTQ, and VCF formats) and phenotypic (including media files) information, using both real and simulated datasets. Advantages of our centralized database concept include the sublinear running time for queries after increasing the number of samples/markers exponentially, in addition to the comprehensive management of distinct data formats while searching for specific genomic regions. A comparison of our non‐relational and generic solution, with an existing relational approach (developed for tabular data types using 2 bits to store genotypes), showed reduced importing time to handle 50M SNPs (PLINK format) achieved by the relational schema. Our experimental results also reinforce that data conversion is a costly step required to manage genomic data into both relational and non‐relational database systems, and therefore, must be carefully treated for large applications.