Two approaches to building models for prediction of the onset of Type 1 diabetes mellitus in juvenile subjects were examined. A set of tests performed immediately before diagnosis was used to build classifiers to predict whether the subject would be diagnosed with juvenile diabetes. A modified training set consisting of differences between test results taken at different times was also used to build classifiers to predict whether a subject would be diagnosed with juvenile diabetes. Neural networks were compared with decision trees and ensembles of both types of classifiers. Support Vector Machines were also tested on this dataset. The highest known predictive accuracy was obtained when the data was encoded to explicitly indicate missing attributes in both cases. In the latter case, high accuracy was achieved without test results which, by themselves, could indicate diabetes. The effects of oversampling of minority class samples in the training set by generating synthetic examples were tested with ensemble techniques like bagging and random forests. It was observed, that oversampling of diabetic examples, lead to an increased accuracy in diabetic prediction demonstrated by a significantly better F-measure value. ROC curves and the statistical F-measure were used to compare the performance of the different machine learning algorithms.

Traditional power system stability assessment based on full model computation shows its drawbacks in real-time applications where fast variations are present at both demand side and supply side. This work presents the use of data mining techniques, in particular the Decision Trees (DTs), for fast evaluation of power system oscillatory stability and voltage stability from synchrophasor measurements. A regression tree-based approach is proposed to predict the stability margins. Modal analysis and continuation power flow are the tools used to build the knowledge base for off-line DT training. Corresponding metrics include the damping ratio of critical electromechanical oscillation mode and MW-distance to the voltage instability region. Classification trees are used to group an operating point into predefined stability state based on the value of corresponding stability indicator. A novel methodology for knowledge base creation has been elaborated to assure practical and sufficient training data. Encouraging results are obtained through performance examination. The robustness of the proposed predictor to measurement errors and system topological variations is analyzed. A scheme has been proposed to tackle the problem of when and how to update the data mining tool for seamless online stability monitoring. The optimal placement for the phasor measurement units (PMU) based on the importance of DT variables is suggested. A measurement-based voltage stability index is proposed and evaluated using field PMU measurements. It is later revised to evaluate the impact of wind generation on distribution system voltage stability. Next, a new data mining tool, the Probabilistic Collocation Method (PCM), is presented as a computationally efficient method to conduct the uncertainty analysis. As compared with the traditional Monte Carlo simulation method, the collocation method could provide a quite accurate approximation with fewer simulation runs. Finally, we show how to overcome the disadvantages of mode meters and ringdown analyzers by using DTs to directly map synchrophasor measurements to predefined oscillatory stability states. The proposed measurement-based approach is examined using synthetic data from simulations on IEEE test systems, and PMU measurements collected from field substations. Results indicate that the proposed method complements the traditional model-based approach, enhancing situational awareness of control center operators in real time stability monitoring and control.

This study investigated the potential of developing in-season crop yield forecasting and mapping systems based on interpretation of airborne hyperspectral remote sensing imagery by machine learning algorithms. The data used for this study was obtained over a corn (Zea mays L.) field in eastern Canada. / The experimental plots were set up at the Emile A. Lods Agronomy Research Center, Montreal, Quebec. Corn was grown under the twelve combinations of three nitrogen application rates (60, 120, and 250 kg N/ha), and four weed control strategies (Broad leaf weed, Grass weed, Broad leaf and grass weed control, and no weed control). The images of the experimental field were taken with a Compact Airborne Spectrographic Imager (CASI) at three times (June 30 for early growth stage, August 5 for tassel stage, and Aug 25 for mature stage) during the year 2000 growing season. / Two machine learning algorithms, Artificial Neural Networks (ANN) and Decision Tree (DT) were evaluated. The performance of ANNs was compared with four conventional modeling methods. For the DT algorithms, two different aspects, (i) DT as a classification method, and (ii) DT as a feature selection tool, were explored in this study.

Diss. (sammanfattning) Stockholm : Karolinska institutet, 2005. / Härtill 4 uppsatser.

Thesis (M.S.)--Oregon State University, 2011. / Printout. Includes bibliographical references (leaves 44-50). Also available on the World Wide Web.

Made available in DSpace on 2015-05-14T12:47:11Z (GMT). No. of bitstreams: 1 arquivototal.pdf: 3933359 bytes, checksum: 8b1a14ddde7e9f62a39543e1408efd75 (MD5) Previous issue date: 2012-02-27 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / Traffic accidents produce high morbidity and mortality in several countries, including Brazil. Initial care to victims of these accidents, by a specialized team, has tools for evaluating severity of trauma, which guide priorities. The purpose of this study is to understand process of decision making to meet victims of traffic accidents and from that develop an understanding of the decision support model that assists medical regulator to decide the severity of injury caused by this condition to health. The study looked at all victims of traffic accidents attended by SAMU of João Pessoa-PB in 2010. It is an epidemiological investigation based on institutional data collection instrument which was the regulation of medical records. Descriptive and spatial statistics was conducted, definition of the decision model was a decision tree whose objective attribute is represented by severity of the injury Abbreviated Injury Scale (AIS). SAMU attended 4.514 TA victims in João Pessoa in 2010. 99% of emergency care to victims were made by teams of basic units. Most victims were male (75.45%), aged between 20 and 39 years (60%) and the mechanism of injury was motorcycle (63%). The most affected body region was limbs (62%) and the more frequent AIS was AIS1 (64.3%). With regard to spatial analysis, the risk map identified the neighborhood Center as the highest risk (10.15) followed by Água Fria (3.23) and Penha (3.15). The spatial scan map that best fitted the risk map was 5% of the population and 5% significance level. The decision model chosen was decision tree that could correctly classify 99.9% of the severity of lesions, with kappa statistics 99.8%. By this model, it was possible to extract 36 rules for classification of the lesion. Given mistakes made by medical regulation on the presumed severity depending on the 192 system information, the use of decision tree makes it possible to reduce subjectivity in decisions to maximize their probability of a hit and consequent reduction in morbidity and mortality brought about by traffic accident. / Os acidentes de trânsito produzem alta morbimortalidade em vários países do mundo, inclusive no Brasil. O atendimento inicial às vítimas destes acidentes, por equipe especializada, conta com instrumentos de avaliação da gravidade do trauma, que norteiam as prioridades. A proposta deste estudo é elaborar um modelo de suporte à decisão que auxilie o profissional médico regulador na tarefa de definir a gravidade da lesão provocada por esse agravo à saúde. No estudo analisaram-se todas as vítimas de acidentes de trânsito atendidas pelo Serviço de Atendimento Médico de Urgência (SAMU) de João Pessoa-PB no ano de 2010. Trata-se de uma investigação epidemiológica baseada em dados institucionais cujo instrumento de coleta foram as fichas de regulação médica. Realizou-se a estatística descritiva, espacial e a definição do modelo de decisão como uma árvore de decisão e cujo atributo objetivo é a gravidade da lesão determinada pela Abbreviated Injury Scale (AIS). O SAMU atendeu 4.514 vítimas de acidentes de trânsito (AT) em João Pessoa no ano de 2010. Verificou-se que 99% desses atendimentos foram realizados por Unidades de Suporte Básico à vida (USB). A maioria das vítimas era do sexo masculino (75,45%), com idade entre 20 e 39 anos (60%) e o mecanismo do trauma foi motocicleta (63%). A região corpórea mais atingida foram os membros (62%) e o AIS mais frequente foi AIS1 (64,3%). Verificou-se também, o envio inadequado de recursos no atendimento às vítimas de AT, tanto USA em vez de USB como o contrário. Com relação à análise espacial, o mapa de risco identificou o bairro centro como sendo o de maior risco (10,1) seguido de Água Fria (3,23) e Penha (3,15). Quanto ao mapa de varredura scan, o que melhor se adequou ao mapa de risco foi a 5% da população e 5% de significância. O modelo de decisão eleito foi a árvore de decisão que classificou corretamente 99,9% das gravidades das lesões, com estatística kappa 99,8%. Por este modelo, foi possível a extração de 36 regras de classificação da lesão. Diante dos equívocos cometidos pelo médico regulador acerca da gravidade presumida, em função das informações transmitidas pelo sistema 192, a utilização da árvore de decisão torna possível a redução da subjetividade nas decisões maximizando sua probabilidade de acerto e a consequente redução da morbimortalidade acarretada pelo acidente de trânsito.

Submitted by Rodrigo Vieira Guidelli null (rguidelli4@gmail.com) on 2016-03-26T16:03:57Z No. of bitstreams: 1 Rodrigo Vieira Guidelli - Dissertação de Mestrado.pdf: 2574087 bytes, checksum: c4570e90e2f553c165e4cbcb47a0f339 (MD5) / Approved for entry into archive by Ana Paula Grisoto (grisotoana@reitoria.unesp.br) on 2016-03-28T16:40:24Z (GMT) No. of bitstreams: 1 guidelli_rv_me_rcla.pdf: 2574087 bytes, checksum: c4570e90e2f553c165e4cbcb47a0f339 (MD5) / Made available in DSpace on 2016-03-28T16:40:24Z (GMT). No. of bitstreams: 1 guidelli_rv_me_rcla.pdf: 2574087 bytes, checksum: c4570e90e2f553c165e4cbcb47a0f339 (MD5) Previous issue date: 2016-02-28 / Pró-Reitoria de Extensão Universitária (PROEX UNESP) / Pró-Reitoria de Pós-Graduação (PROPG UNESP) / Em 2009 Uehara Prado et al., coletaram uma grande quantidade de dados para avaliar o papel das borboletas da família Nymphalidae como bioindicadoras, porém esses dados não foram utilizados em sua totalidade. O presente estudo está direcionado à experimentação e modelagem de interações ecológicas, a partir dos dados obtidos por Uehara-Prado et al. (2009), juntamente com aqueles não previamente utilizados que, no intuito de extrair o máximo de informação de relevância biológica e ecológica. Para tanto, foram utilizados três diferentes tipos de abordagens: (1) Biclusterização (Cheng & Church, 2000; Madeira & Oliveira, 2004); (2) Árvores de decisão (Quinlan, 1986; Bell, 1999; De’ath & Fabricius, 2000; Olden et al., 2008) e (3) Redes Bayesianas (Korb & Nicholson, 2003; McCann et al., 2006; Chen & Pollino, 2012; Pearl, 2014). Os resultados se mostraram bastante promissores, e as três ferramentas atingiram as expectativas; em biclusterização, conseguimos identificar todos os padrões de correlação dentro dos cenários apresentados, árvores de decisão se mostraram extremamente eficazes na classificação das variáveis apresentadas e as Redes Bayesianas conseguiram identificar quais variáveis influenciavam ou eram influenciadas pelas outras. Com este trabalho esperamos incentivar outros pesquisadores à revisitarem antigas bases de dados com ferramentas computacionais mais modernas, pois seu potencial é extraordinário. / Elucidating the complex interactions networks in ecological systems is not an easy task (Proulx et al., 2005) and, in order to extract information in an efficient way, powerful computational tools and the right approach, to the types of scenario to be studied, are required. In 2009 UeharaPrado et al., collected a great amount of data to assess the role of the Nymphalidae family of butterflies as bio-indicators, but these data were not used in its entirety. This study is aimed at experimentation and modeling of the ecological interactions from the data obtained by UeharaPrado et al. (2009), along with those not previously used, in order to extract the maximum information of biological and ecological significance. Therefore, three different approaches were used: (1) Biclusterization (Cheng & Church, 2000; Wood& Olive, 2004); (2) Decision Trees (Quinlan, 1986; Bell, 1999; De'ath & Fabricius., 2000; Olden et al, 2008) and (3) Bayesian Networks (Korb & Nicholson, 2003; McCann et al., 2006; Chen & Pollino, 2012; Pearl, 2014). The results were very promising, and the three tools reached our expectations; with Biclusterization we managed to identify all the correlation patterns inside the scenarios presented, Decision Trees proved to be extremely effective in the classification of the variables and the Bayesian Networks were able to identify what variables influenced or were influenced by the others. With this work, we hope to encourage other researchers to revisit old databases with more modern computational tools, because its potential is extraordinary.

Interest in distributed approaches to machine learning has increased significantly in recent years due to continuously increasing data sizes for training machine learning models. In this thesis we describe three popular machine learning algorithms: decision trees, Naive Bayes and support vector machines (SVM) and present existing ways of distributing them. We also perform experiments with decision trees distributed with bagging, boosting and hard data partitioning and evaluate them in terms of performance measures such as accuracy, F1 score and execution time. Our experiments show that the execution time of bagging and boosting increase linearly with the number of workers, and that boosting performs significantly better than bagging and hard data partitioning in terms of F1 score. The hard data partitioning algorithm works well for large datasets where the execution time decrease as the number of workers increase without any significant loss in accuracy or F1 score, while the algorithm performs poorly on small data with an increase in execution time and loss in accuracy and F1 score when the number of workers increase.

This thesis investigates the issue of the application of cognitive analysis techniques for Western art music to folk dance melodies for violin, with a view to enabling the development of a computer tool that can aid in the identification and exploration of the stylistic characteristics of the origin of the melodies. The following questions are addressed: Can cognitive music analysis techniques for Western art music be applied successfully to folk dance melodies for violin? Is it possible to define an integrative analysis approach in this context drawing from existing approaches? To what extent can decision tree induction aid in the classification and interpretation of the analysis results? How might the musical data for analysis be represented on computer? What is the best approach to program development for an automated music analysis tool in this context? A series of experiments using samples of American and Irish melodies are presented that verify the use, in this context, of the cognitive analysis approaches of Lerdahl and lackendoff and Narmour. Statistical approaches have also been investigated, since research has shown that such methods can reflect the way in which listeners mentally organise the music that they hear. To enable the analysis to be carried out in an algorithmic way, an experiment using human subjects to further the work of Lerdahl and lackendoff was required. An integrative analysis approach has been identified that can be carried out in an algorithmic way therefore lending itself to future implementation on computer. In order to interpret the results of the analysis process, a decision tree induction tool (SeeS) based on Quinlan's CS algorithm was employed. SeeS was able to classify the melodies based on the attributes derived from the analysis. The decision trees and rules derived by the tool enabled the identification of features of the melodies that pertain to their origins, thus enabling a deeper understanding of the stylistic variations of the melodies. A further experiment indicated that the cognitive analysis approaches and subsequent classification with SeeS compares favourably with the classification abilities of human subjects after a small amount of training in the musical context. Further inductive learning techniques (decision tree induction using Friedman's CART, and neural networks) have been applied to the problem of classification andinterpretation of the analysis results, and although the neural network classified the musical samples with greater accuracy (illustrated using ROC analysis), decision tree induction has been shown to be a more appropriate method in this context. Approaches to music representation and subsequent program development have been investigated, reSUlting in a proposal for future computer implementation of a music analysis tool using the Humdrum toolkit as a means of representation, and a declarative language for the program development.

Data mining approaches have been increasingly used in recent years in order to find patterns and regularities in large databases. In this study, the C4.5 decision tree approach was used for mining of Gene Logic database, containing biological data. The decision tree approach was used in order to identify the most relevant genes and risk factors involved in breast cancer, in order to separate healthy patients from breast cancer patients in the data sets used. Four different tests were performed for this purpose. Cross validation was performed, for each of the four tests, in order to evaluate the capacity of the decision tree approaches in correctly classifying ‘new’ samples. In the first test, the expression of 108 breast related genes, shown in appendix A, for 75 patients were used as input to the C4.5 algorithm. This test resulted in a decision tree containing only four genes considered to be the most relevant in order to correctly classify patients. Cross validation indicates an average accuracy of 89% in classifying ‘new’ samples. In the second test, risk factor data was used as input. The cross validation result shows an average accuracy of 87% in classifying ‘new’ samples. In the third test, both gene expression data and risk factor data were put together as one input. The cross validation procedure for this approach again indicates an average accuracy of 87% in classifying ‘new’ samples. In the final test, the C4.5 algorithm was used in order to indicate possible signalling pathways involving the four genes identified by the decision tree based on only gene expression data. In some of cases, the C4.5 algorithm found trees suggesting pathways which are supported by the breast cancer literature. Since not all pathways involving the four putative breast cancer genes are known yet, the other suggested pathways should be further analyzed in order to increase their credibility. In summary, this study demonstrates the application of decision tree approaches for the identification of genes and risk factors relevant for the classification of breast cancer patients