<p>It is rarely possible to use an optimal classifier. Often the classifier used for a specific problem is an approximation of the optimal classifier. Methods are presented for evaluating the performance of an approximation in the model class of Bayesian Networks. Specifically for the approximation of class conditional independence a bound for the performance is sharpened.</p><p>The class conditional independence approximation is connected to the minimum description length principle (MDL), which is connected to Jeffreys’ prior through commonly used assumptions. One algorithm for unsupervised classification is presented and compared against other unsupervised classifiers on three data sets.</p> / Report code: LiU-TEK-LIC 2006:11.

La présente thèse envisage de cerner les comportements des Camerounais sur les routes.En l'occurrence, nous voulons savoir comment les croyances fatalistes, les croyances decontrôle, les croyances au contrôle divin, les croyances et valeurs culturelles, l'explicationnaïve des accidents et la perception du risque peuvent affecter les comportements des usagerset agents du trafic routier au Cameroun. Cinq études sont réalisées dans le cadre de notrethèse. La première étude examine 522 procès-verbaux d'accidents réels et montre qu'unegrande majorité des accidents surviennent dans de bonnes conditions de conduite. De plus, lesconducteurs se rejettent mutuellement la faute, mais s'accordent avec les gendarmes pour direque le comportement des conducteurs est la première cause des accidents de la route auCameroun. Une deuxième étude évalue la variation des comportements en fonction des explicationscausales et des croyances, sur un échantillon de 525 participants. On observe que lesparticipants présentent davantage des comportements sécuritaires lorsqu'ils expliquent lesaccidents par le comportement des conducteurs, mais leurs comportements tendent à êtremoins sécuritaires lorsqu'ils expliquent ceux-ci par des causes externes non contrôlables. Lesexplications causales tendent à être externes aux conducteurs lorsque les participants sontfatalistes, croient qu'ils peuvent affronter les situations de trafic dangereuses sans en êtreinquiété, croient que Dieu contrôle les situations dangereuses auxquelles ils peuvent faire faceou lorsqu'ils adhèrent fortement aux croyances et valeurs culturelles. En particulier, lescroyances et valeurs culturelles qui sont supposées protéger la vie, les croyances fatalistes etles croyances au contrôle divin se distinguent par leur capacité à favoriser l'explication desaccidents par des forces invisibles et à induire des comportements insécuritaires. Enfin, le rôlemédiateur des explications causales se révèlent pour toutes les croyances.A partir d'une quasi-expérimentation réalisée auprès de 444 participants, l'étude 3analyse la variation des explications causales et de l'attitude vis-à-vis des mesures deprévention, en fonction de la pertinence situationnelle, de la pertinence personnelle et de lagravité de l'accident. On observe que les participants ont tendance à fournir des explicationscausales défensives d'autant plus que la situation accidentelle leur est pertinente, qu'ilss'identifient à la victime et que l'accident est grave. De plus ils ont une préférence pour desmesures de prévention orientées vers les conducteurs lorsqu'ils expliquent les accidents par lecomportement de ces derniers. L'étude 4 montre une tendance à adopter des comportements moins sécuritaires lorsque les participants sous-estiment le risque routier. En outre, ces derniers ont tendance à sousestimer le risque lorsqu'ils sont fatalistes alors qu'ils ont davantage peur du risque lorsqu'ils sont attachés à leur identité culturelle. Enfin, les participants ont tendance à se croire capables d'affronter le risque routier sans en être inquiété lorsqu'ils croient que Dieu contrôle lessituations dangereuses ou lorsqu'ils croient aux pratiques culturelles supposées protéger lavie. Dans l'étude 5, on montre que les participants ont tendance à adopter descomportements davantage sécuritaires lorsqu'ils ont une perception élevée du risque etexpliquent les accidents par des causes contrôlables. Par contre, ils se montrent plutôtimprudents sur les routes lorsqu'ils ont une perception élevée du risque et croient que lesaccidents sont causés par des forces invisibles. Les résultats vont dans le sens des travaux antérieurs et sont discutés en rapport avec les connaissances théoriques. Enfin, des suggestions encouragent une prévention fondée sur les croyances de la population cible.

In this dissertation we developed theoretical models to analyse Gaussian and multinomial distributions. The analysis is focused on classification in high dimensional feature spaces and provides a basis for dealing with issues such as data sparsity and feature selection (for Gaussian and multinomial distributions, two frequently used models for high dimensional applications). A Naïve Bayesian philosophy is followed to deal with issues associated with the curse of dimensionality. The core treatment on Gaussian and multinomial models consists of finding analytical expressions for classification error performances. Exact analytical expressions were found for calculating error rates of binary class systems with Gaussian features of arbitrary dimensionality and using any type of quadratic decision boundary (except for degenerate paraboloidal boundaries). Similarly, computationally inexpensive (and approximate) analytical error rate expressions were derived for classifiers with multinomial models. Additional issues with regards to the curse of dimensionality that are specific to multinomial models (feature sparsity) were dealt with and tested on a text-based language identification problem for all eleven official languages of South Africa. / Thesis (M.Ing. (Computer Engineering))--North-West University, Potchefstroom Campus, 2009.

This dissertation explored the way in which bumblebees' visual system helps them discover their first flower. Previous studies found bees have unlearned preferences for parts of a flower, such as its colour and shape. The first study pitted two variables against each other: pattern type: sunburst or bull's eye, versus the location of the pattern: shapes appeared peripherally or centrally. We observed free-flying bees in a flight cage using Radio-Frequency Identification (RFID) tracking. The results show two distinct behavioural preferences: Pattern type predicts landing: bees prefer radial over concentric patterns, regardless of whether the radial pattern is on the perimeter or near the centre of the flower. Pattern location predicts exploration: bees were more likely to explore the inside of artificial flowers if the shapes were displayed near the centre of the flower, regardless of whether the pattern was radial or concentric. As part of the second component, we implemented a mathematical model aimed at explaining how bees come to prefer radial patterns, leafy backgrounds and symmetry. The model was based on unsupervised neural networks used to describe cognitive mechanisms. The results captured with the results of multiple behavioural experiments. The model suggests that bees choose computationally "cheaper" stimuli, those that contain less information. The third study tested the computational load hypothesis generated by the artificial neural networks. Visual properties of symmetry, and spatial frequency were tested. Studying free-flying bees in a flight cage using motion-sensitive video recordings, we found that bees preferred 4-axis symmetrical patterns in both low and high frequency displays.

In this dissertation we developed theoretical models to analyse Gaussian and multinomial distributions. The analysis is focused on classification in high dimensional feature spaces and provides a basis for dealing with issues such as data sparsity and feature selection (for Gaussian and multinomial distributions, two frequently used models for high dimensional applications). A Naïve Bayesian philosophy is followed to deal with issues associated with the curse of dimensionality. The core treatment on Gaussian and multinomial models consists of finding analytical expressions for classification error performances. Exact analytical expressions were found for calculating error rates of binary class systems with Gaussian features of arbitrary dimensionality and using any type of quadratic decision boundary (except for degenerate paraboloidal boundaries). Similarly, computationally inexpensive (and approximate) analytical error rate expressions were derived for classifiers with multinomial models. Additional issues with regards to the curse of dimensionality that are specific to multinomial models (feature sparsity) were dealt with and tested on a text-based language identification problem for all eleven official languages of South Africa. / Thesis (M.Ing. (Computer Engineering))--North-West University, Potchefstroom Campus, 2009.

A travers la présente thèse, nous nous intéressons à la prévention des accidents d’avalanche survenant à l’occasion de la pratique du hors-piste. Les jeunes pratiquants du hors-piste (15-30 ans) sont repérés comme les principales victimes des avalanches par les données statistiques de l’Association Nationale pour l’Etude de la Neige et des Avalanches. L’objectif principal de la présente thèse est de parvenir à une meilleure connaissance de ce public adepte de sports de glisse. Pour cela, nous proposons d’investiguer leurs habitudes et motivations de pratique du hors-piste, leurs perceptions des risques liés à cette pratique, les explications qu’ils donnent spontanément pour les accidents d’avalanche en hors-piste, ainsi que les comportements qu’ils adoptent en hors-piste et les critères qu’ils utilisent pour prendre la décision de sortir des pistes. Les données recueillies devraient contribuer à l’enrichissement des stratégies de prévention des accidents d’avalanche en hors-piste. Elles permettent de concevoir des stratégies et des messages ciblés et adaptés à cette population de jeunes pratiquants. En effet, les travaux sur la perception des risques (Kouabenan, 2006) et sur l’explication naïve des accidents (Kouabenan, 1999) suggèrent que si l’on souhaite que les individus adhèrent aux messages de prévention, il est nécessaire de connaître la façon dont ils appréhendent les risques, et pour ce qui nous concerne les risques en hors-piste, notamment le risque d’avalanche. Il importe aussi de connaître les causes qu’ils invoquent pour expliquer la production des accidents d’avalanche. Ce travail de thèse comprend quatre études dont la méthodologie repose sur deux questionnaires. L’élaboration des questionnaires s’appuie sur des entretiens préalables menés avec des pratiquants aux profils divers, et sur l’exploitation de la littérature sur les travaux empiriques empruntant le même cadre théorique ainsi que sur les rapports sur les accidents d’avalanche. Concernant les études 1, 2 et 4, nous avons interrogé 304 jeunes pratiquants de sports de glisse. Pour l’étude 4, nous avons questionné 238 pratiquants. Pour le recueil des données concernant les quatre études, nous avons rencontré les participants sur le terrain, dans des stations de sports d’hiver en Isère, en Savoie et en Haute-Savoie, ainsi que sur le campus universitaire et lors de manifestations sportives. Le recueil des données s’est fait par entretien en face à face avec le participant. / Through this thesis, we examine the avalanche accident prevention occurring in out-of-bounds (OB) practice. Young out-of-bounds practitioners (15-30 years-old) are identified as the core of avalanche victims. The main aim of this thesis is to get better knowledge about these boardsports enthusiasts. To do this, we suggest to investigate their habits and motivations relative to OB practice, their perception of the risks associated to this practice, the spontaneous explanations they give for avalanche accidents in OB, as well as the behaviours they adopt in OB, and the criteria they use to decide to go or not to go OB. The data collected would contribute to improve prevention strategies for avalanche accidents in OB, by targeting and adapting these strategies to young practitioners. Indeed, research on risk perception (Kouabenan, 2006) and naive causal explanations of accidents (Kouabenan, 1999) propose that if we want people to adhere to prevention messages, it is necessary to know the way they understand risks, for our case OB risks, particularly avalanche hazard. It is also essential to know the causes they invoke to explain avalanche accidents. This thesis work consists of four studies for which the methodology relies on two questionnaires. The questionnaires were carried out from preliminary interviews conducted with practitioners of varied profiles, and from literature and empirical works of the same theoretical framework, as well as accidents reports. Concerning the studies 1, 2 and 4, we interviewed 304 young boardsports practitioners. For study 3, we questioned 238 practitioners. For the data collect of the four studies, we met participants in the field, in winter sport resorts in Isère, Savoie and Haute-Savoie (France), as well as at the university campus and during sport events. Data collection was done with face-to-face interview with the participant.

A presente dissertação analisa o conteúdo dos diários e cartas de Virginia Woolf para avaliar se um algoritmo de classificação de texto poderia identificar um padrão escrito relacionado aos dois meses anteriores ao suicídio de Virginia Woolf. Este é um estudo de classificação de texto. Comparamos 46 entradas de textos dos dois meses anteriores ao suicídio de Virginia Woolf com 54 textos selecionados aleatoriamente do trabalho de Virginia Woolf durante outro período de sua vida. O texto de cartas e dos diários foi incluído, enquanto livros, romances, histórias curtas e fragmentos de artigos foram excluídos. Os dados foram analisados usando um algoritmo de aprendizagem mecânica Naïve-Bayes. O modelo mostrou uma acurácia de 80,45%, sensibilidade de 69% e especificidade de 91%. A estatística Kappa foi de 0,6, o que significa um bom acordo, e o valor P do modelo foi de 0,003. A Área Sob a curva ROC foi 0,80. O presente estudo foi o primeiro a analisar a viabilidade de um modelo de machine learning, juntamente com dados de texto, a fim de identificar padrões escritos associados ao comportamento suicida nos diários e cartas de um romancista. Nossa assinatura de texto foi capaz de identificar o período de dois meses antes do suicídio com uma alta precisão / The present study analyzes the content of Virginia Woolf’s diaries and letters to assess whether a text classification algorithm could identify written pattern related to the two months previous to Virginia Woolf’s suicide. This is a text classification study. We compared 46 texts entries from the two months previous to Virginia Woolf’s suicide with 54 texts randomly selected from Virginia Woolf’s work during other period of her life. Letters and diaries were included, while books, novels, short stories, and article fragments were excluded. The data was analyzed by using a Naïve-Bayes machine-learning algorithm. The model showed a balanced accuracy of 80.45%, sensitivity of 69%, and specificity of 91%. The Kappa statistic was 0.6, which means a good agreement, and the p value of the model was 0.003. The Area Under the ROC curve was 0.80. The present study was the first to analyze the feasibility of a machine learning model coupled with text data in order to identify written patterns associated with suicidal behavior in the diaries and letters of a novelist. Our text signature was able to identify the period of two months preceding suicide with a high accuracy.

Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / This present study aims ay analyzing the contribuition of abolicionists from the State of Sergipe for the project of the nation s civilization, through practices of instruction and educational to naïve children. This research, delimited between the ages of 1881 to 1884, studies the educational laws of the imperial period and the aspects focused on and related to the primary instruction. It unveils the motives of the creation of an elementary teaching school and its pedagogical practices for the primary teaching to the naïve children capita of the Sergipe settlement destined to children as well as the pedagogical practices developed by the school. It identifies the profile of the Sociedade Libertadora Aracajuana Cabana do Pai Thomaz (1882) as a space for speeches, conferences, journalistic activities, delivery of emancipation letters and instruction to naïve children. It has also focused in this investiment, the role of the press and the usages made by its editors at the and of the nineteenth century. The theoreticmethodological for the study in question is supported in Roger Chartier, throught the notions of Representation an Appropriation, the concepts of Civilization by Norbert Elias, Cultura Escolar by Dominique Julia, Infância by Carlos Monarcha and the category Constructos Flutuantes by Clarice Nunes, Therefore, it treats about a study which focus on the set of pratices carried o9n by abolicionists from Sergipe for reaching goals of progress in Sergipe. / O presente estudo tem como objetivo analisar a contribuição de abolicionistas sergipanos para o projeto de civilização da nação, através das práticas de instrução e educação aos ingênuos. Esta pesquisa, delimitada entre os anos de 1881 a 1884 estuda as leis educacionais do império e os aspectos abordados no que diz respeito a instrução primária Desvenda os motivos da criação de uma escola de ensino primário e suas práticas pedagógicas para o ensino primário aos ingênuos, na Capital da Província Sergipana destinada aos ingênuos bem como, as práticas pedagógicas desenvolvidas pela escola. Identifica o perfil da Sociedade Libertadora Aracajuana Cabana do Pai Thomaz (1882) como um espaço de palestras, conferências, atividades jornalísticas, entregas de cartas de alforria e instrução aos ingênuos. Pôs em relevo também nesta investida, o papel da imprensa e os usos feitos pelos seus redatores no final do século XIX.. O referencial teórico-metodológico para o estudo em questão apóia-se em Roger Chartier, a partir das noções de Representação e Apropriação, os conceitos de Civilização de Norbert Elias, Cultura escolar de Dominique Julia, Infância de Carlos Monarcha e a categoria Constructos flutuantes de Clarice Nunes. Trata-se, portanto, de um estudo que focaliza o conjunto de práticas exercidas por abolicionistas sergipanos para alcançar metas de progresso em Sergipe.

A presente dissertação analisa o conteúdo dos diários e cartas de Virginia Woolf para avaliar se um algoritmo de classificação de texto poderia identificar um padrão escrito relacionado aos dois meses anteriores ao suicídio de Virginia Woolf. Este é um estudo de classificação de texto. Comparamos 46 entradas de textos dos dois meses anteriores ao suicídio de Virginia Woolf com 54 textos selecionados aleatoriamente do trabalho de Virginia Woolf durante outro período de sua vida. O texto de cartas e dos diários foi incluído, enquanto livros, romances, histórias curtas e fragmentos de artigos foram excluídos. Os dados foram analisados usando um algoritmo de aprendizagem mecânica Naïve-Bayes. O modelo mostrou uma acurácia de 80,45%, sensibilidade de 69% e especificidade de 91%. A estatística Kappa foi de 0,6, o que significa um bom acordo, e o valor P do modelo foi de 0,003. A Área Sob a curva ROC foi 0,80. O presente estudo foi o primeiro a analisar a viabilidade de um modelo de machine learning, juntamente com dados de texto, a fim de identificar padrões escritos associados ao comportamento suicida nos diários e cartas de um romancista. Nossa assinatura de texto foi capaz de identificar o período de dois meses antes do suicídio com uma alta precisão / The present study analyzes the content of Virginia Woolf’s diaries and letters to assess whether a text classification algorithm could identify written pattern related to the two months previous to Virginia Woolf’s suicide. This is a text classification study. We compared 46 texts entries from the two months previous to Virginia Woolf’s suicide with 54 texts randomly selected from Virginia Woolf’s work during other period of her life. Letters and diaries were included, while books, novels, short stories, and article fragments were excluded. The data was analyzed by using a Naïve-Bayes machine-learning algorithm. The model showed a balanced accuracy of 80.45%, sensitivity of 69%, and specificity of 91%. The Kappa statistic was 0.6, which means a good agreement, and the p value of the model was 0.003. The Area Under the ROC curve was 0.80. The present study was the first to analyze the feasibility of a machine learning model coupled with text data in order to identify written patterns associated with suicidal behavior in the diaries and letters of a novelist. Our text signature was able to identify the period of two months preceding suicide with a high accuracy.

Data Ductus, a multination IT-consulting company, wants to develop an AI that monitors a logistic system and looks for errors. Once trained enough, this AI will suggest a correction and automatically right issues if they arise. This project presents how one works with machine learning problems and provides a deeper insight into how cross-validation and regularisation, among other techniques, are used to improve the performance of machine learning algorithms on the defined problem. Three techniques are tested and evaluated in our logistic system on three different machine learning algorithms, namely Naïve Bayes, Logistic Regression and Random Forest. The evaluation of the algorithms leads us to conclude that Random Forest, using cross-validated parameters, gives the best performance on our specific problems, with the other two falling behind in each tested category. It became clear to us that cross-validation is a simple, yet powerful tool for increasing the performance of machine learning algorithms. / Data Ductus, ett multinationellt IT-konsultföretag vill utveckla en AI som övervakar ett logistiksystem och uppmärksammar fel. När denna AI är tillräckligt upplärd ska den föreslå korrigering eller automatiskt korrigera problem som uppstår. Detta projekt presenterar hur man arbetar med maskininlärningsproblem och ger en djupare inblick i hur kors-validering och regularisering, bland andra tekniker, används för att förbättra prestandan av maskininlärningsalgoritmer på det definierade problemet. Dessa tekniker testas och utvärderas i vårt logistiksystem på tre olika maskininlärnings algoritmer, nämligen Naïve Bayes, Logistic Regression och Random Forest. Utvärderingen av algoritmerna leder oss till att slutsatsen är att Random Forest, som använder korsvaliderade parametrar, ger bästa prestanda på våra specifika problem, medan de andra två faller bakom i varje testad kategori. Det blev klart för oss att kors-validering är ett enkelt, men kraftfullt verktyg för att öka prestanda hos maskininlärningsalgoritmer.