23º SINAPE - Simpósio Nacional de Probabilidade e Estatística

Página Inicial » Inscrições Científicas » Trabalhos

Dados do Trabalho


Título

RECENT ADVANCES IN TIME SERIES ANALYSIS AND PATTERN RECOGNITION

Resumo Geral da Sessão Temática

In the modern world applications, time series analysis represents one of the most important topics in statistics, as data is being continuously generated in industry (e.g. sensor data, anomaly detection), social media (online searches, video/image/post comments) and finances (e.g. stock markets, GDP growth), among others. With the increase in the amount of data being collected, challenges such as computational time, data contamination and high dimensionality often arise. In this thematic session we will present a general overview and recent advances for some of the time series methodologies, both parametric and non-parametric. Applications to industry, business and finances are considered.

Nome do Palestrante 1 e Moderador / Instituição / Currículo / Título da Fala / Resumo da fala

Nome: Paulo Canas Rodrigues

Instituição: Universidade Federal da Bahia (UFBA)

CV Lattes: http://lattes.cnpq.br/0029960374321970

Título: Singular spectrum analysis for long and contaminated time series

Resumo: Singular spectrum analysis (SSA) is a non-parametric method for time series analysis and forecasting that incorporates elements of classical time series analysis, multivariate statistics, multivariate geometry, dynamical systems and signal processing. Although this technique has shown to be advantageous over traditional model based methods, in particular, one of the steps of the SSA algorithm, which refers to the singular value decomposition (SVD) of the trajectory matrix, is highly sensitive to data contamination and also time consuming.
In this talk we will present (i) the randomized SSA which is an alternative to SSA for long time series without losing the quality of the analysis; and (ii) a robust SSA algorithm, where a robust SVD procedure replaces the least-squares based SVD in the original SSA procedure, in order to reduce the effect of data contamination by outlying observations.
The SSA and the randomized SSA are compared in terms of quality of the model fit and forecasting accuracy, and computational time, via Monte Carlo simulations and real data about the daily prices of five of the major world commodities. The SSA and the robust SSA are compared in terms of the quality of the model fit via Monte Carlo simulations that contemplate both clean and noisy/contaminated time series, and considering a real data application where a data set from the energy sector is analyzed.

Nome do Palestrante 2 / Instituição / Currículo / Título da Fala / Resumo da fala

Nome: Ricardo Ehlers

Instituição: ICMC/USP

CV Lattes: http://lattes.cnpq.br/4020997206928882

Título: Hamiltonian Monte Carlo methods in financial time series

Resumo: In this work we develop and apply novel Markov chain Monte Carlo methods to estimate parameters and compare models for financial time series of returns. We employ a Bayesian approach where the posterior distributions are approximated using Hamiltonian Monte Carlo (HMC), an optimized version of the Metropolis-Hasting algorithm. The zero variance principle is also exploited to find unbiased
estimators with smaller variances. The ideas presented are assessed using both simulated and real data
of financial returns.
Keywords: Bayesian inference, Hamiltonian Monte Carlo, influential points, zero variance MCMC.

Nome do Palestrante 3 / Instituição / Currículo / Título da Fala / Resumo da fala

Nome: Thelma Sáfadi

Instituição: Departamento de Estatística - UFLA

CV Lattes: http://lattes.cnpq.br/9821585244827807

Título: Sunflower seeds diagnostics via 2-D Wavelet-based Spectral Descriptors

Resumo: Analysis of seeds is essential for determining seed lot quality and hence its sowing value. Although subjective by its interpretation, X-ray images of seeds are an important media for analyzing seed lot quality and has been used as alternative to standard laboratory testing. However, evaluation of radiographic images still depends on the analyst’s experience. This subjectivity may be reduced with techniques for automatic processing of images in which software-driven analysis helps analysts in the evaluation process. Considering the 2-D discrete non-decimated wavelet transform two approaches are presented. In the first we applied 2-D scale-mixing wavelet transform for automatic processing of radiographic images. From the transformed images several spectral indices are derived. These descriptors involve spectral slopes which are directly connected with the degree of image regularity. In the second the classification of the x-ray images of seeds as damaged or undamaged are considered based on the directional Hurst exponents. In this case, two location measures (mean and median) are considered.

Palavras-Chave

Time series
Singular spectrum analysis
Hamiltonian Monte Carlo
Adaptive LASSO

Público alvo

Researchers and students interested in (or curious about) time series analysis.

Justificativa

Being time series analysis one of the most important topics in statistical applications nowadays, the dissemination of its current developments is of great importance for SINAPE and for the Brazilian Statistical community, both researchers and students.

Recursos

Os palestrantes nesta sessão irão procurar ativamente fontes de recursos que garantam o financiamento da viagem, estadia e taxa de inscrição.

Outras informações

Área

Geral

Autores

Paulo Canas Rodrigues, Thelma Sáfadi, Ricardo Ehlers