The Workshop on Statistics and Data Science 2024 is organized by CMAT-SAPOR from the Centre of Mathematics at the University of Minho. It focuses on topics in Statistics, Applied Probability, and Operational Research, and is held as part of the celebration of the ninth European Statistics Day, which is officially celebrated on October 20th.
For the course, the participants should bring their laptops and have installed the following free softwares: Python ou Anaconda e PyCharm. Due to the physical capacity of the laboratory, the course is limited to 50 participants.

With this workshop, we intend to promote the exchange of ideas and collaborations between
researchers from the University of Minho and the wider scientific community, providing
opportunities to interact with experts across various fields

Invited speakers:

• Adelaide Freitas
• Ana Moreira
• Cecília Martins
• Daniel Tinoco
• F. Catarina Pereira
• Luís Meira-Machado
• Maria Eduarda Silva
• Rosemeire Leovigildo Fiaccone
• Soraia Pereira

Local Organizing Committee:

• A. Manuela Gonçalves
• Inês Sousa
• Luis Machado
• Maria Conceição Serra
• Marta Ferreira
• Raquel Menezes
• Susana Faria

    2012
    2013
    2014
    2019
    2021
    2022

Program

Thursday, October 24th 2024

Short Course Bruno Veloso Assistant Professor at FEP - University of Porto and Senior Researcher at INESC TEC Knowledge Discovery from Data Streams (KD-DS) Knowledge Discovery from Data Streams (KD-DS) Abstract: Knowledge Discovery from Data Streams (KD-DS) refers to the process of extracting valuable insights and patterns from continuously flowing data in real-time. With the rapid proliferation of sensors, social media, financial transactions, and IoT devices, vast amounts of data are generated at an unprecedented pace, making traditional batch processing methods inefficient for handling such dynamic environments. KD-DS addresses this challenge by employing algorithms and techniques designed for incremental learning, adaptability, and real-time analysis. Core tasks include stream classification, clustering, regression, and anomaly detection, often under constraints of limited memory, processing power, and the need for timely results. Techniques like sliding windows, data summarization, and concept drift detection are essential to manage evolving patterns over time. KD-DS has found applications in various fields, including cybersecurity, financial monitoring, and smart city infrastructure, where real-time decision-making is crucial.

09:30-10:30

Part I

10:30-11:00

Coffee Break

11:00-12:30

Part II

12:30-14:00

Lunch

14:00-15:30

Part III

15:30-16:00

Coffee Break

16:00-17:00

Part IV

Friday, October 25th 2024

09:15-09:45

Luís Meira-Machado - President of the Portuguese Statistical Society European Statistics Day European Statistics Day Abstract: The opening session of the Workshop on Statistics and Data Science 2024 will be led by the President of the Portuguese Statistical Society. In addition to offering a warm welcome to all participants and officially launching the event, the session will highlight the significance of the European Statistics Day, celebrated on October 20th. This day underscores the vital role of statistics in society, aligning with the workshop's focus on advancing knowledge in Statistics, Applied Probability, and Operational Research.

09:45-10:25

Adelaide Freitas - Department of Mathematics and Center for Research and Development in Mathematics and Applications (CIDMA), University of Aveiro, Portugal Performance of CDPCA Performance of CDPCA Abstract: Clustering and Disjoint Principal Component Analysis (CDPCA) is a constrained Principal Component Analysis aimed at a simultaneous clustering of objects along a set of centroids and a partitioning of variables along a set of sparse and disjoint components. In this talk, we focus our discussion on the performance of CDPCA on some datasets when the Alternating Least Square algorithm is used in the estimation of the parameter of the CDPCA model (a joint work with Maurizio Vichi and Márcia Sartori).

10:25-11:05

Coffee Break

11:05-11:30

Daniel Tinoco- PhD program in Mathematics, Center of Mathematics, University of Minho Distributed Generalized Linear Models Distributed Generalized Linear Models Abstract: This presentation will introduce an innovative approach to classical linear regression, allowing model computation in distributed settings with privacy-preserving measures for federated data. It'll also extend the methodology to encompass a generalized linear model that maintains these properties while accommodating more diverse data distributions.

11:30-11:55

F. Catarina Pereira, A. Manuela Gonçalves, and Marco Costa - MAP-PDMA - PhD in Applied Mathematics, Centre of Mathematics, Department of Mathematics, University of Minho Improving Weather Forecasts with State-Space Models: Predicting Maximum Temperature Improving Weather Forecasts with State-Space Models: Predicting Maximum Temperature Abstract: Climate change has increased the frequency of droughts, affecting water management, especially in agriculture [1]. This work focuses on improving short-term forecasts of maximum temperature, a key variable in the evapotranspiration process. By using state-space models and the Kalman filter, the accuracy of 1 to 6-day ahead forecasts has been improved [2]. This presentation will explain how these models work and demonstrate their potential for real-time weather prediction.

11:55-12:20

Ana Moreira and Susana Faria - PhD program in Mathematics, Centre of Mathematics, Department of Mathematics, University of Minho L1-penalization approach for variable selection in mixtures of linear regression models L1-penalization approach for variable selection in mixtures of linear regression models Abstract: Finite Mixture Regression (FMR) models provide a flexible tool for modelling data that arise from a heterogeneous population, where the relationship between the dependent and explanatory variables varies across different subpopulations. Given the often large number of explanatory variables in such models, variable selection becomes critically important. Traditional methods can be computationally demanding, so penalty-based methods like the Least Absolute Shrinkage and Selection Operator (LASSO), Adaptive LASSO (ALASSO), and Relaxed LASSO (RLASSO) have been developed. This study compares the performance of these methods in selecting the most relevant explanatory variables in mixtures of linear regression models. Extensive simulation analyses demonstrate that the ALASSO method shows superior overall performance.

12:20-14:15

Lunch

14:15-14:55

Maria Eduarda Silva - Faculty of Economics, University of Porto and LIADD INESC TEC Nowcasting macroeconomic indicators from online activity data: a case study with unemployment rates Nowcasting macroeconomic indicators from online activity data: a case study with unemployment rates Abstract: Data from Google Trends (GT) have been gaining importance as predictors for economic indicators to overcome delays allowing timely forecasting. Such data have emerged in the literature as alternative predictors of macroeconomic outcomes, such as the unemployment rate, featuring readiness, public availability and no costs. This talk introduces extensive daily GT data to develop a framework to nowcast monthly unemployment rates in a real-time data availability environment, resorting to Mixed Data Sampling (MIDAS) regressions. Portugal is chosen as a use case for the methodology since the extraction of GT data requires the selection of keywords, which is culture-dependent. The nowcasting period comprises from 2019 to 2021, which includes the coronavirus outbreak. The results show that daily GT predictors via MIDAS lead to accurate and timely information on the unemployment rate and are particularly effective in dealing with the external shock from COVID-19, showing accuracy gains even when compared to nowcasts obtained from typical monthly GT data via traditional ARIMAX models.

14:55-15:35

Soraia Pereira - Centre of Statistics and its Applications of University of Lisbon (CEAUL), Department of Mathematics, University of Minho A Hierarchical Bayesian Geostatistical Model for Spatial Data with Excess Zeros and Extreme Values: Analyzing Sardine Egg Density in Portugal A Hierarchical Bayesian Geostatistical Model for Spatial Data with Excess Zeros and Extreme Values: Analyzing Sardine Egg Density in Portugal Abstract: Modeling spatial datasets characterized by an excess of zeros and extreme values poses significant challenges in statistical analysis. Such data structures are common in ecological studies, particularly in marine biology, where the presence or absence of a species and rare high-density occurrences can impact population management strategies. This study introduces a hierarchical Bayesian geostatistical model designed to effectively handle these complexities in the context of sardine egg density along the Portuguese coast.

15:35-16:15

Coffee Break

16:15-16:55

Rosemeire Leovigildo Fiaccone - Professora Titular, Departamento de Estatística, Instituto de Matemática e Estatística, Universidade Federal da Bahia Statistical modeling applications in the health field Statistical modeling applications in the health field Abstract: Count data are present in the most diverse research areas and are usually analyzed through the traditional Poisson and Negative Binomial regression models. There are other possible models depending on the research question and the behavior of this type of data. The idea is to show some statistical modeling techniques for count data through Bell probability distribution, which has the properties of being uniparametric and accommodating the phenomenon of overdispersion, which is commonly observed in this type of data

16:55-17:20

Cecília Martins - PhD program in Mathematics, Center of Mathematics, University of Minho Modelling Data with Fractional Differential Equations Using Neural Networks Modelling Data with Fractional Differential Equations Using Neural Networks Abstract: Fractional Differential Equations (FDEs) are essential tools for modelling complex systems in science and engineering, enabling a more precise representation of processes characterised by non-local and memory-dependent behaviours. We propose the Neural FDE, a novel deep neural network framework that adjusts a FDE to the dynamics of data. Due to the time-dependence, Neural FDE can be used for irregularly-sampled data unlike traditional neural networks. The numerical results suggest that, despite being more computationally demanding, it can effectively be applied to learn complex dynamical systems.

17:20

Closing