In the last years, the use of machine learning methods has increased remarkably and therefore the research in this field is becoming more and more important. Despite this fact, a high uncertainity when using machine learning models is still present. We have a wide variety of machine learning approaches such as decision trees or support vector machines and many applications where machine learning has been proved useful like medical diagnosis or computer vision, but all this possibilities make finding the best machine learning approach for a given application a time consuming and not welldefined process since there is not a rule that tells us what method to use for a given type of data.We attempt to build a system that, using machine learning, is capable to learn the best machine learning approach for a given application. For that, we are working on the hypothesis that similar types of data will have also the same machine learning approachas best learner. Classification algorithms will be the main focus of this research and different statistical measures will be used in order to find these similarities among the data.

Made available in DSpace on 2014-08-20T14:36:56Z (GMT). No. of bitstreams: 1 Tese_ Alvaro_ Luiz_ Carvalho_ Nebel.pdf: 1615608 bytes, checksum: cf5d12660d9ae82ddc2e973b92c9812e (MD5) Previous issue date: 2009-04-24 / The understanding of the dynamics of the water in the soil-plant-atmosphere system, including the water availability to the crops, soil water infiltration, drainage and soil solute movement, depends on the knowledge of the relation between the soil water content and the matric potential, represented by the soil water retention curve (SWRC). However, the establishment of SWRCs is laborious and time consuming, besides being costly. A alternative is its estimate through statistical equations called Pedotransfer Functions (PTFs). The aim of this study was to evaluate the ability of some existing PTFs, in predicting the soil water retention and to capture its spatial variability structure, using geostatistical tools, when applied in a lowland soil of the south region of Brazil. For this, an experimental 10 x 10 m grid was established and soil disturbed and undisturbed samples were collected in the 0-0.20 m soil depth, totaling 100 experimental points. The following soil attributes were determined in each point: soil texture, soil organic carbon, pH, cation exchange capacity, soil bulk density, and the soil water retention curve. Eight developed PTFs for estimating gravimetric soil water content, eight for estimating volumetric water content and five for estimating the van Genuchten model parameters were evaluated using the statistical measures mean error (ME), and the root mean square error (RMSE). Results indicated that the Oliveira et al. (2002) PTF presented the best performance for estimating the gravimetric soil water content at the tension of 33kPa, with mean error (ME) value of 0.0136g.g-1, while for the gravimetric water content at tension of 1500kPa the Pidgeon (1972) FPT was the best, with ME value of -0,0054g.g-1. Concerning to the potential of describing the spatial variability structure the Bell & van Keulen (1995 and 1996) and of Urach (2007) PTFs presented the best performance based on the results from the cross validation technique. For estimating the soil water volumetric content at the tension of 10kPa the Tomasela et al. (2002) was the best (ME of 0.019cm3.cm-3), while Rawls et al. (1982) and van den Berg et al. (1997) PTFs were the best for estimating soil water content at the tension of 33kPa (ME of 0.001cm3.cm-3) and 1500kPa (ME of -0.008cm3.cm-3), respectively. The range and sill geostatistical parameters for the tension of 33kPa were reasonable estimated, while for the tension of 1500kPa they were underestimated by the evaluated PTFs. The Parametric Pedotransfer Function constructed by Vereecken et al.(1989) presented the lowest value of ME (0.0247cm3.cm-3), while the Hodnett and Tomasella (2002) PTF presented the lowest value of RMSE (0.0367 cm3.cm-3). Both PTFs well described the experimental semivariograms of the soil water content at the tensions of 10kPa and 33kPa, however their performance was not good for the tension of 1500kPa. / Estudos que envolvem a dinâmica da água no sistema solo-planta-atmosfera tais como disponibilidade de água no solo para as culturas, infiltração, drenagem e movimento de solutos no solo necessitam do conhecimento da relação entre o conteúdo de água no solo e o potencial matricial, representada pela curva de retenção de água no solo. No entanto, sua execução é laboriosa, demanda considerável tempo e custos. Uma alternativa é sua estimativa através de equações estatísticas denominadas Funções de Pedotransferência (FPTs). O objetivo deste estudo foi avaliar o desempenho de funções de pedotransferência em estimar a retenção de água no solo e capturar a sua estrutura de variabilidade espacial, usando ferramentas geoestatísticas, quando aplicadas em um solo de várzea da região sul do Brasil. Para isto, uma malha experimental de 10m x 10m foi estabelecido e amostras deformadas e indeformadas do solo foram coletadas representativas da camada de 0 0,20m, totalizando 100 pontos amostrais. Os seguintes atributos do solo foram determinados em cada ponto: textura, carbono orgânico, pH, capacidade de troca de cátions, densidade do solo e a curva de retenção de água no solo. Oito FPTs desenvolvidas para estimar o conteúdo gravimétrico de água no solo, oito para estimar o conteúdo volumétrico e cinco para estimar os parâmetros do modelo de van Genuchten foram avaliadas, usando as medidas estatísticas erro médio (ME) e raiz quadrada do erro médio ao quadrado (RMSE). Os resultados obtidos para a estimativa do conteúdo de água gravimétrico mostram que a FPT de Oliveira et al. (2002) apresentou o melhor desempenho para a tensão de 33kPa, com erro médio (ME) de 0,0136g.g-1, enquanto para a tensão de 1500kPa a FPT de Pidgeon (1972) foi melhor, com ME de -0,0054g.g-1. Com relação ao potencial de descrever a estrutura de variabilidade espacial do conjunto de dados medidos, as FPTs desenvolvidas por Bell & van Keulen (1995 e 1996) e Urach (2007) foram as que apresentaram melhor desempenho baseado nos resultados da validação cruzada. Para o conteúdo volumétrico de água retida no solo, os melhores desempenhos foram obtidos pela FPT de Tomasela et al.(2003) para a tensão de 10kPa, Rawls et al. (1982) para 33kPa e van den Berg et al.(1997) para 1500kPa, com ME iguais a 0,019cm3.cm-3, 0,001cm3.cm-3 e -0,008cm3.cm-3, respectivamente. Os parâmetros da estrutura de dependência espacial, alcance e patamar, para a tensão de 33kPa foram razoavelmente estimados, enquanto para a tensão de 1500kPa foram subestimados pelas FPTs avaliadas. A FPT paramétrica desenvolvida por Vereecken et al. (1989) apresentou o menor valor de ME (0,0247cm3.cm-3), enquanto a FPT de Hodnett e Tomasella (2002) apresentou o menor valor de RMSE (0,0367 cm3.cm-3). Ambas FPTs descreveram bem os semivariogramas experimentais do conteúdo de água retido no solo a 10kPa e 33kPa, entretanto para a tensão de 1500kPa nenhuma das FPTs testadas apresentou bom desempenho.

No description available.