Anselmo Chaves Neto

Applying correlation to enhance boosting technique using genetic programming as base learner

This paper explores the Genetic Programming and Boosting technique to obtain an ensemble of regressors and proposes a new formula for the updating of weights, as well as for the final hypothesis. Differently from studies found in the literature, in this paper we investigate the use of the correlation metric as an additional factor for the error metric. This new approach, called Boosting using Correlation Coefficients (BCC) has been empirically obtained after trying to improve the results of the other methods. To validate this method, we conducted two groups of experiments. In the first group, we explore the BCC for time series forecasting, in academic series and in a widespread Monte Carlo simulation covering the entire ARMA spectrum. The Genetic Programming (GP) is used as a base learner and the mean squared error (MSE) has been used to compare the accuracy of the proposed method against the results obtained by GP, GP using traditional boosting and the traditional statistical methodology (ARMA). The second group of experiments aims at evaluating the proposed method on multivariate regression problems by choosing Cart (Classification and Regression Tree) as the base learner.

O uso de quartis para a aplicação dos filtros de Graham na Bovespa (1998-2009)

With a view to advancing in the understanding of the Brazilian stock market and analyzing appropriate strategies for small investors, this article suggest a review on the application of Grahams filters to stock selection in Bovespa between 1998 and 2009. Using a portfolio approach and statistical tests the presence of returns above normal and higher than Ibovespa returns was noticed s one, especially for five-year portfolios. However, these portfolios had low diversification of assets. As Graham filters were developed for the U.S. market in the 70s, the development of new qualifiers was suggested for the sake of adaptation to the current Brazilian scenario. This approach used the most interesting quartiles of each criterion to define the new qualifiers and showed a return higher than Ibovespa for all periods analyzed, but did not eliminate the problem of low portfolio diversification.

Time series forecasting with the WARIMAX-GARCH method

It is well-known that causal forecasting methods that include appropriately chosen Exogenous Variables (EVs) very often present improved forecasting performances over univariate methods. However, in practice, EVs are usually difficult to obtain and in many cases are not available at all. In this paper, a new causal forecasting approach, called Wavelet Auto-Regressive Integrated Moving Average with eXogenous variables and Generalized Auto-Regressive Conditional Heteroscedasticity (WARIMAX-GARCH) method, is proposed to improve predictive performance and accuracy but also to address, at least in part, the problem of unavailable EVs. Basically, the WARIMAX-GARCH method obtains Wavelet EVs (WEVs) from Auto-Regressive Integrated Moving Average with eXogenous variables and Generalized Auto-Regressive Conditional Heteroscedasticity (ARIMAX-GARCH) models applied to Wavelet Components (WCs) that are initially determined from the underlying time series. The WEVs are, in fact, treated by the WARIMAX-GARCH method as if they were conventional EVs. Similarly to GARCH and ARIMA-GARCH models, the WARIMAX-GARCH method is suitable for time series exhibiting non-linear characteristics such as conditional variance that depends on past values of observed data. However, unlike those, it can explicitly model frequency domain patterns in the series to help improve predictive performance. An application to a daily time series of dam displacement in Brazil shows the WARIMAX-GARCH method to remarkably outperform the ARIMA-GARCH method, as well as the (multi-layer perceptron) Artificial Neural Network (ANN) and its wavelet version referred to as Wavelet Artificial Neural Network (WANN) as in [1], on statistical measures for both in-sample and out-of-sample forecasting.

Using Correlation to Improve Boosting Technique: An Application for Time Series Forecasting

Time series forecasting has been widely used to support decision making, in this context a highly accurate prediction is essential to ensure the quality of the decisions. Ensembles of machines currently receive a lot of attention; they combine predictions from different forecasting methods as a procedure to improve the accuracy. This paper explores genetic programming and boosting technique to obtain an ensemble of regressors and proposes a new formula for the final hypothesis. This new formula is based on the correlation coefficient instead of the geometric median used by the boosting algorithm. To validate this method, experiments were performed, the mean squared error (MSE) has been used to compare the accuracy of the proposed method against the results obtained by GP, GP using a boosting technique and the traditional statistical methodology (ARMA). The results show advantages in the use of the proposed approach

A influência das atividades mineradoras na alteração do pH e da alcalinidade em águas fluviais: o exemplo do rio Capivari, região do carste paranaense

Mining activities causes many kinds of environmental impacts over environment. In this paper we describe the research developed in the Capivari watershed, in the Curitiba metropolitan region, Parana, Brazil. Alkalinity and pH changes in Capivari stream over 12 years were observed. The sampling program was performed in 387 sampling days series at 1986/1987, and another period covering 1,095 days from 1998 to 2000. Comparing these series, the pH average was higher in 0,5 unities and the alkalinity were 15% higher within this period of time. From 1980 to 2001, mining areas of metadolomites spread out to an average rate of 47,000 m2/year. Field evaluations carried out on the Capivari stream showed that in regions close to low water streams and closer to stone-pits, the pH level was higher and it became more alkaline as the electric conductivity increased. These data support the hypotheses that the expansion of mining areas would affect the pH and the alkalinity of the Capivari stream.

Itaipu Hydroelectric Power Plant Structural Geotechnical Instrumentation Temporal Data Under the Application of Multivariate Analysis – Grouping and Ranking Techniques

© 2012 Villwock et al., licensee InTech. This is an open access chapter distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Itaipu Hydroelectric Power Plant Structural Geotechnical Instrumentation Temporal Data Under the Application of Multivariate Analysis – Grouping and Ranking Techniques

Model-Based Vibration Condition Monitoring for Fault Detection and Diagnostics in Large Hydrogenerators

Large hydrogenerators are rotating machines of vertical assembly, equipped with tilting pad journal bearings, operating at subcritical speeds (80 to 200 rpm). The nominal power of these machines may reach 700 MW, what makes vibration-based condition monitoring a compulsory task. Several international standards are applicable to vibration monitoring of hydrogenerators. However, their condition assessment criteria are grounded on measurements performed on a wide set of hydrogenerators, with power varying from 1 to 700 MW and speed in the range 60 to 1800 rpm. As result, some limits are excessively tight while others are too much permissive. Moreover, experimental observations revealed that the dynamic behavior of these machines might present significant short and long-term changes, many times with no apparent reasons. Part of this behavior is due to the lack of a defined radial static load in the journal bearings, as well as to external agents, like generator electromagnetic field or seasonal variations of bearing cooling water temperature. All these aspects make difficult the using of statistical pattern recognition. It is necessary a better understanding of the influencing mechanisms of the vibratory behavior of these machines, to differentiate normal changes from those originated by incipient faults. This paper proposes the using of a model-based approach to overcome these problems and exemplifies this using on a set of 700 MW hydrogenerators. The results obtained indicated that even simplified models might present satisfactory results, especially when models performance are improved using additional information collected by the monitoring system.

ESTIMAÇÃO DO GRAU DE ASTIGMATISMO PELO MÉTODO SUPPORT VECTOR REGRESSION CORRELACIONADO

DOI: 10.12957/cadest.2016.27523 Este trabalho apresenta a influencia do coeficiente de correlacao em modelos de regressao. Para o desenvolvimento, utilizou-se o metodo Support Vector Regression (SVR) como modelo de regressao. O metodo SVR foi aplicado em exames feitos em pacientes que possuem algum nivel de astigmatismo. Para tanto, criou-se uma modificacao na fase de ajuste do modelo de regressao, sendo introduzido o coeficiente de correlacao linear, avaliando a correlacao entre as variaveis preditoras: Ceratometria, subdividida em Eixo Mais Plano e Eixo Mais Curvo; e Refracao, subdividido em Esfera e Cilindrico, sendo o grau de astigmatismo a variavel a ser prevista. O novo metodo proposto, nomeado SVR Correlacionado teve seus resultados comparados com o metodo convencional SVR, obtendo um desempenho superior, tanto na correlacao dos modelos como no valor do erro cometido. Ao todo, utilizaram-se os dados de 26 pacientes com astigmatismo, sendo criadas duas configuracoes para o ajuste e teste, a primeira sendo composta de 20 observacoes para ajuste e seis para teste apresentando o erro RMSE = 0,035799, e a segunda composta de 16 observacoes na fase de ajuste e 10 no teste, gerando RMSE = 0,028518, em ambos os casos, inferiores aos erros gerados pelo metodo SVR convencional.

Factor analysis in data mining applied for recognition and classification pattern for smart grid

The present work aims to present the method of factor analysis applied to pre-processing of data mining process. The factor analysis is applied to two pattern recognition and classifications databases of power engineerings problems. This pre-processing analyzes the input variables, thus having a better understanding of the importance of each variable as a input for pattern recognition and classifications methods. The pattern recognition and classifications methods were applied to the data with and without the application of factor analysis, thus comparing the real rate of right classification.

Quality assessment of insulating oil in transformers with application of quadratic discriminant function

The mineral oil insulating power transformers is subject to deterioration due to mechanical and chemical conditions of use. In the course of use, the oil is subjected to oxidation reactions due to the presence of oxygen, water, and metals. Monitoring and maintaining the quality of the insulating oil are essential steps to ensure the reliability in operation of the transformers. The main objective of this paper is to present a methodology of Pattern Recognition based on Multivariate Statistical Technique, which identify the quality of the insulating oil how good quality, able to be recovered or to be discarded. This decision is made from the vector of chemical analysis of the basic characteristics of the oil. The automation of this rule of recognition and classification is proposed in this work by means of a computer program in C++ language compatible with the systems of electricity distribution companies.