As more and more remotely sensed data becomes available it is becoming increasingly harder to analyse it with the more traditional labour intensive, manual methods. The commonly used techniques, that involve expert evaluation, are widely acknowledged as providing inconsistent results, at best. We need more general techniques that can adapt to a given situation and that incorporate the strengths of the traditional methods, human operators and new technologies. The difficulty in interpreting remotely sensed data is that often only a small amount of data is available for classification. It can be noisy, incomplete or contain irrelevant information. Given that the training data may be limited we demonstrate a variety of techniques for highlighting information in the available data and how to select the most relevant information for a given classification task. We show that more consistent results between the training data and an entire image can be obtained, and how misclassification errors can be reduced. Specifically, a new technique for attribute selection in neural networks is demonstrated. Machine learning techniques, in particular, provide us with a means of automating classification using training data from a variety of data sources, including remotely sensed data and expert knowledge. A classification framework is presented in this thesis that can be used with any classifier and any available data. While this was developed in the context of vegetation mapping from remotely sensed data using machine learning classifiers, it is a general technique that can be applied to any domain. The emphasis of the applicability for this framework being domains that have inadequate training data available.

As more and more remotely sensed data becomes available it is becoming increasingly harder to analyse it with the more traditional labour intensive, manual methods. The commonly used techniques, that involve expert evaluation, are widely acknowledged as providing inconsistent results, at best. We need more general techniques that can adapt to a given situation and that incorporate the strengths of the traditional methods, human operators and new technologies. The difficulty in interpreting remotely sensed data is that often only a small amount of data is available for classification. It can be noisy, incomplete or contain irrelevant information. Given that the training data may be limited we demonstrate a variety of techniques for highlighting information in the available data and how to select the most relevant information for a given classification task. We show that more consistent results between the training data and an entire image can be obtained, and how misclassification errors can be reduced. Specifically, a new technique for attribute selection in neural networks is demonstrated. Machine learning techniques, in particular, provide us with a means of automating classification using training data from a variety of data sources, including remotely sensed data and expert knowledge. A classification framework is presented in this thesis that can be used with any classifier and any available data. While this was developed in the context of vegetation mapping from remotely sensed data using machine learning classifiers, it is a general technique that can be applied to any domain. The emphasis of the applicability for this framework being domains that have inadequate training data available.

Thèse numérisée par la Division de la gestion de documents et des archives de l'Université de Montréal.

Thèse numérisée par la Division de la gestion de documents et des archives de l'Université de Montréal

Submitted by Marcos Luciano Santos (marcosluciano@gmail.com) on 2016-02-27T19:29:54Z No. of bitstreams: 1 DISSERTACAO-MLOS-2016.pdf: 762474 bytes, checksum: 35e4816c58bc9bfc0ed334edec259e5b (MD5) / Approved for entry into archive by Renata de Souza Nascimento (renata.souza@fgv.br) on 2016-02-29T16:44:53Z (GMT) No. of bitstreams: 1 DISSERTACAO-MLOS-2016.pdf: 762474 bytes, checksum: 35e4816c58bc9bfc0ed334edec259e5b (MD5) / Made available in DSpace on 2016-02-29T16:56:14Z (GMT). No. of bitstreams: 1 DISSERTACAO-MLOS-2016.pdf: 762474 bytes, checksum: 35e4816c58bc9bfc0ed334edec259e5b (MD5) Previous issue date: 2016-02-02 / This paper aims to study the closure of multi-strategy investment funds in the Brazilian market considering the influence of the variables: management and performance fees, age, fund performance, net equity and fund flow. The selected funds are open-end and public distribution, and studied in two different observation windows, annual and quarterly, from 2008 to 2014. Two types of funds were studied: Master and investment funds in quotas. To determine their performance two different metrics were used: return discounted DI rate and Sharpe ratio. The variables were evaluated under two different models, the first using variables’ data as absolute values and the second classifying them into quintiles. The results were obtained by applying regressions in logistics panel, panel and cross section. The results showed the importance of the equity net and the fund flow, followed by age. The performances, return discounted DI rate and the Sharpe ratio, were less influential. The other variables, management fee and performance fee, were not significant. A second result is the difference of the variables significance between the fund types: Master and FIC. Among Master funds, net equity had consistent influence on closing events in annual observations. Another important finding is the difference between quarterly and annual windows, many variables especially the performance ones have a higher level of significance in shorter windows. / Este trabalho tem por objetivo o estudo sobre o encerramento de fundos de investimento multiestratégia no mercado brasileiro com relação à influência das variáveis: taxas de administração e performance, idade, desempenho, patrimônio e captação líquida. Os fundos selecionados apresentam características de condomínio aberto, e destinados a investidores em geral, com duas janelas de observação, anuais e trimestrais, durante o período de 2008 a 2014. Os estudos apresentados são separados para dois tipos de fundos: com classificação Master e fundos de investimento em cotas (FIC). Para a determinação de desempenho dos fundos utiliza-se duas diferentes métricas, o retorno descontado da taxa DI e o índice de Sharpe. As variáveis foram avaliadas sob dois distintos modelos, o primeiro utilizando os dados das variáveis em valores absolutos e o segundo classificando-as em quintis. Para a obtenção dos resultados os dados foram aplicados por regressões em painel logística, painel e cross section. Os resultados obtidos mostraram a importância do patrimônio e da captação líquida, seguido por idade. Os desempenhos, tanto retorno descontado quanto o índice de Sharpe, apresentaram menor influência, e com pouca significância os custos de administração e de performance. Um segundo resultado apresentado é a diferença da significância das variáveis entre os tipos de fundos Master e FIC, no grupo de fundos Master apenas patrimônio líquido apresentou consistente influência nos eventos de encerramento em janelas anuais. Outra importante constatação é a diferença existente entre janelas trimestrais e anuais, em que grande parte das variáveis estudadas, principalmente as variáveis de desempenho apresentam maior nível de significância nas janelas mais curtas.

Continuous change and development are essential aspects of evolving environments and applications, including, but not limited to, smart cities, military, medicine, nuclear reactors, self-driving cars, aviation, and aerospace. That is, the fundamental characteristics of such environments may evolve, and so cause dangerous consequences, e.g., putting people lives at stake, if no reaction is adopted. Therefore, learning systems need to apply intelligent algorithms to monitor evolvement in their environments and update themselves effectively. Further, we may experience fluctuations regarding the performance of learning algorithms due to the nature of incoming data as it continuously evolves. That is, the current efficient learning approach may become deprecated after a change in data or environment. Hence, the question 'how to have an efficient learning algorithm over time against evolving data?' has to be addressed. In this thesis, we have made two contributions to settle the challenges described above. In the machine learning literature, the phenomenon of (distributional) change in data is known as concept drift. Concept drift may shift decision boundaries, and cause a decline in accuracy. Learning algorithms, indeed, have to detect concept drift in evolving data streams and replace their predictive models accordingly. To address this challenge, adaptive learners have been devised which may utilize drift detection methods to locate the drift points in dynamic and changing data streams. A drift detection method able to discover the drift points quickly, with the lowest false positive and false negative rates, is preferred. False positive refers to incorrectly alarming for concept drift, and false negative refers to not alarming for concept drift. In this thesis, we introduce three algorithms, called as the Fast Hoeffding Drift Detection Method (FHDDM), the Stacking Fast Hoeffding Drift Detection Method (FHDDMS), and the McDiarmid Drift Detection Methods (MDDMs), for detecting drift points with the minimum delay, false positive, and false negative rates. FHDDM is a sliding window-based algorithm and applies Hoeffding’s inequality (Hoeffding, 1963) to detect concept drift. FHDDM slides its window over the prediction results, which are either 1 (for a correct prediction) or 0 (for a wrong prediction). Meanwhile, it compares the mean of elements inside the window with the maximum mean observed so far; subsequently, a significant difference between the two means, upper-bounded by the Hoeffding inequality, indicates the occurrence of concept drift. The FHDDMS extends the FHDDM algorithm by sliding multiple windows over its entries for a better drift detection regarding the detection delay and false negative rate. In contrast to FHDDM/S, the MDDM variants assign weights to their entries, i.e., higher weights are associated with the most recent entries in the sliding window, for faster detection of concept drift. The rationale is that recent examples reflect the ongoing situation adequately. Then, by putting higher weights on the latest entries, we may detect concept drift quickly. An MDDM algorithm bounds the difference between the weighted mean of elements in the sliding window and the maximum weighted mean seen so far, using McDiarmid’s inequality (McDiarmid, 1989). Eventually, it alarms for concept drift once a significant difference is experienced. We experimentally show that FHDDM/S and MDDMs outperform the state-of-the-art by representing promising results in terms of the adaptation and classification measures. Due to the evolving nature of data streams, the performance of an adaptive learner, which is defined by the classification, adaptation, and resource consumption measures, may fluctuate over time. In fact, a learning algorithm, in the form of a (classifier, detector) pair, may present a significant performance before a concept drift point, but not after. We define this problem by the question 'how can we ensure that an efficient classifier-detector pair is present at any time in an evolving environment?' To answer this, we have developed the Tornado framework which runs various kinds of learning algorithms simultaneously against evolving data streams. Each algorithm incrementally and independently trains a predictive model and updates the statistics of its drift detector. Meanwhile, our framework monitors the (classifier, detector) pairs, and recommends the efficient one, concerning the classification, adaptation, and resource consumption performance, to the user. We further define the holistic CAR measure that integrates the classification, adaptation, and resource consumption measures for evaluating the performance of adaptive learning algorithms. Our experiments confirm that the most efficient algorithm may differ over time because of the developing and evolving nature of data streams.

People often have very little knowledge about the impact of unsustainable human population growth on the environment and social well-being especially in developing countries. Therefore, an efficient method should be explored in order to educate, and if possible, to convince the members of the public to realize the environmental and social problems caused by the unsustainable population growth. Digital Game-Based Learning (DGBL) has been highlighted by some studies as an innovative tool for learning enhancement. While only a handful of studies have scientifically evaluated the impact of DGBL on knowledge outcomes, the approach is an attractive tool to increase knowledge and motivate engagement with environmental issues surrounding population growth because of its potential to improve learners’ motivation and engagement thereby compared to traditional learning approaches. Therefore, the three primary research questions for this study are: 1) "Can a single-player digital game be an appropriate and attractive learning application for the players to gain insight about the relationship between the growing human population and the environmental issues?" 2) "How can we design environmental games for the players to gain insights about the relationship between the growing human population and the environmental issues via playing a game?" and 3) "What are the obstacles preventing the players from adapting environmental knowledge obtained from the learning mediums into the real-life?" To inform the development of an efficacious DGBL game to impact learning outcomes, critical reviews of environmental issues related to population growth as well as critical reviews of commercial and serious environmental games in terms of their educational and motivational values were undertaken in this study. The results of these critical reviews informed the development of a Guideline for Environmental Games (or GEG). The GEG was developed by combining the engaging game technology with environmental learning and persuasion theories. The GEG was then used to inform the development of a prototype game called THE GROWTH; a single-player, quiz-based, city-management game targeting young adolescents and adults. Multiple evaluation methods of the game were used to answer the three key research questions mentioned earlier. These methods included: 1) The Randomized Controlled Trial approach (RCT) where the participants were systematically divided into the experimental and the control group respectively and their knowledge scores (quantitative data) compared and analyzed, 2) The participants’ abilities to recall and describe the environmental and well-being issues were collected and analyzed qualitatively using The Content Analysis method (CA) and, 3) The participants’ overall feedback on the learning mediums was collected and analyzed to evaluate the motivational values of THE GROWTH itself. To this end, THE GROWTH was evaluated with 82 Thai-nationality participants (70 males and 12 females). The results showed that participants assigned to play THE GROWTH demonstrated greater environmental and social-well-being knowledge related to population growth (F(1,40) = 43.86, p = .006) compared to the control group participants assigned to a non-interactive reading activity (consistent with material presented in THE GROWTH). Furthermore, participants who played THE GROWTH recalled on average more content presented in the game when compared to participants who were presented with similar content in the reading material (t (59) = 3.35, p = .001). In terms of level of engagement, the study suggested that participants assigned to the game were more engaging with their learning medium on average when compared to participants assigned to the non-interactive reading activity. This is evidenced by the longer time participants spent on the task, the activity observed from participants’ recorded gameplay, and their positive responses in the survey. The semi-structured interviews used in this study highlighted the participants’ attitudes towards the environmental, social, and technological issues. Although the participants’ perceived behavioural intention towards the environmental commitments were not statistically differed between the two study group, their responses still provide some evidences that leaps may occur from the learning mediums to the real-world context. Furthermore, these responses can be valuable evidences for the policy makers and for the future development of environmental serious games. Overall, the results suggested that digital environmental games such as THE GROWTH might be an effective and motivational tool in promote the learning about sustainable population size, the environment, and the social well-being. The game’s ability to convince the participants to change towards sustainable lifestyles, however, might be subjected to the future research and other real-world circumstances such as the governmental and public supports. In summary, the research in this thesis makes the following contributions to knowledge: • The Guideline for Environmental Games (GEG) contributes to knowledge about making theoretically-based environmental games. It has particular significance because the guideline was validated by demonstrating learning improvements in a systematic randomized controlled trial. • The use of Multi-Strategy Study Design where multiple systematic evaluation methods were used in conjunction to provide conclusive findings about the efficacy of DGBL to impact outcomes. • THE GROWTH itself is a contribution to applied research as an example of an effective DGBL learning tool.