Em anos recentes, muitos algoritmos bio-inspirados têm surgido para resolver problemas de classificação. Em confirmação a isso, a revista Nature, em 2002, publicou um artigo que já apontava para o ano de 2003 o uso comercial de Sistemas Imunológicos Artificiais para detecção de fraude em instituições financeiras por uma empresa britânica. Apesar disso, não observamos, a luz de nosso conhecimento, nenhuma publicação científica com resultados promissores desde então. Nosso trabalho tratou de aplicar Sistemas Imunológicos Artificiais (AIS) para detecção de fraude em cartões de crédito. Comparamos AIS com os métodos de Árvore de Decisão (DT), Redes Neurais (NN), Redes Bayesianas (BN) e Naive Bayes (NB). Para uma comparação mais justa entre os métodos, busca exaustiva e algoritmo genético (GA) foram utilizados para selecionar um conjunto paramétrico otimizado, no sentido de minimizar o custo de fraude na base de dados de cartões de crédito cedida por um emissor de cartões de crédito brasileiro. Em adição à essa otimização, fizemos também uma análise e busca por parâmetros mais robustos via multi-resolução, estes parâmetros são apresentados neste trabalho. Especificidades de bases de fraude como desbalanceamento de dados e o diferente custo entre falso positivo e negativo foram levadas em conta. Todas as execuções foram realizadas no Weka, um software público e Open Source, e sempre foram utilizadas bases de teste para validação dos classificadores. Os resultados obtidos são consistentes com Maes et al. que mostra que BN são melhores que NN e, embora NN seja um dos métodos mais utilizados hoje, para nossa base de dados e nossas implementações, encontra-se entre os piores métodos. Apesar do resultado pobre usando parâmetros default, AIS obteve o melhor resultado com os parâmetros otimizados pelo GA, o que levou DT e AIS a apresentarem os melhores e mais robustos resultados entre todos os métodos testados. / In 2002, January the 31st, the famous journal Nature, with a strong impact in the scientific environment, published some news about immune based systems. Among the different considered applications, we can find detection of fraudulent financial transactions. One can find there the possibility of a commercial use of such system as close as 2003, in a British company. In spite of that, we do not know of any scientific publication that uses Artificial Immune Systems in financial fraud detection. This work reports results very satisfactory on the application of Artificial Immune Systems (AIS) to credit card fraud detection. In fact, scientific financial fraud detection publications are quite rare, as point out Phua et al. [PLSG05], in particular for credit card transactions. Phua et al. points out the fact that no public database of financial fraud transactions is available for public tests as the main cause of such a small number of publications. Two of the most important publications in this subject that report results about their implementations are the prized Maes (2000), that compares Neural Networks and Bayesian Networks in credit card fraud detection, with a favored result for Bayesian Networks and Stolfo et al. (1997), that proposed the method AdaCost. This thesis joins both these works and publishes results in credit card fraud detection. Moreover, in spite the non availability of Maes data and implementations, we reproduce the results of their and amplify the set of comparisons in such a way to compare the methods Neural Networks, Bayesian Networks, and also Artificial Immune Systems, Decision Trees, and even the simple Naïve Bayes. We reproduce in certain way the results of Stolfo et al. (1997) when we verify that the usage of a cost sensitive meta-heuristics, in fact generalized from the generalization done from the AdaBoost to the AdaCost, applied to several tested methods substantially improves it performance for all methods, but Naive Bayes. Our analysis took into account the skewed nature of the dataset, as well as the need of a parametric adjustment, sometimes through the usage of genetic algorithms, in order to obtain the best results from each compared method.

Introduits au début des années 1990, les systèmes immunitaires artificiels visent à adapter les propriétés du système immunitaire biologique, telles que sa scalabilité et son adaptivité, à des problèmes informatiques : sécurité, mais également optimisation et classification. Cette thèse explore une nouvelle direction en se concentrant non sur les processus biologiques et les cellules elles-mêmes, mais sur les interactions entre les sous-systèmes. Ces modes d’interaction engendrent les propriétés reconnues du système immunitaire : détection d’anomalies, reconnaissance des pathogènes connus, réaction rapide après une exposition secondaire et tolérance à des organismes symbiotiques étrangers. Un ensemble de systèmes en interaction formant un écosystème, cette nouvelle approche porte le nom d’Écosystème Immunitaire Artificiel. Ce modèle est mis à l’épreuve dans un contexte particulièrement sensible à la scalabilité et à la performance : la supervision de réseaux, qui nécessite l’analyse de séries temporelles en temps réel avec un expert dans la boucle, c’est-à-dire en utilisant un apprentissage actif plutôt que supervisé. / Since the early 1990s, immune-inspired algorithms have tried to adapt the properties of the biological immune system to various computer science problems, not only in computer security but also in optimization and classification. This work explores a different direction for artificial immune systems, focussing on the interaction between subsystems rather than the biological processes involved in each one. These patterns of interaction in turn create the properties expected from immune systems, namely their ability to detect anomalies, memorize their signature to react quickly upon secondary exposure, and remain tolerant to symbiotic foreign organisms such as the intestinal fauna. We refer to a set of interacting systems as an ecosystem, thus this new approach has called the Artificial Immune Ecosystem. We demonstrate this model in the context of a real-world problem where scalability and performance are essential: network monitoring. This entails time series analysis in real time with an expert in the loop, i.e. active learning instead of supervised learning.

Submitted by Geandra Rodrigues (geandrar@gmail.com) on 2018-01-08T10:33:23Z No. of bitstreams: 1 tiagofayerdejesus.pdf: 1430604 bytes, checksum: 44c933cea80c72bd24709a956ba5c9cd (MD5) / Approved for entry into archive by Adriana Oliveira (adriana.oliveira@ufjf.edu.br) on 2018-01-22T18:34:11Z (GMT) No. of bitstreams: 1 tiagofayerdejesus.pdf: 1430604 bytes, checksum: 44c933cea80c72bd24709a956ba5c9cd (MD5) / Made available in DSpace on 2018-01-22T18:34:11Z (GMT). No. of bitstreams: 1 tiagofayerdejesus.pdf: 1430604 bytes, checksum: 44c933cea80c72bd24709a956ba5c9cd (MD5) Previous issue date: 2017-08-31 / Neste trabalho é proposta uma metodologia para resolução de problemas de planejamento de redes de distribuição de energia elétrica considerando incertezas na demanda. A metodologia é baseada na técnica meta-heurística Sistema Imunológico Artificial. No processo de solução, a matemática intervalar é usada no fluxo de potência, e nele estão modeladas as incertezas da carga. As variáveis intervalares de entrada são as cargas ativas e reativas nas barras do sistema. O resultado esperado é apresentado de maneira intervalar e consiste na minimização dos custos de instalação de cabos, de construção/reforço de subestações mais os custos associados às perdas de energia e operação considerando um horizonte de planejamento. Para isso, restrições como radialidade, conectividade, balanço de potência e capacidades físicas de equipamentos foram utilizadas. Para determinar a melhor topologia de rede, outra metodologia foi adotada para se fazer a comparação dos intervalos. O objetivo principal deste trabalho é avaliar o impacto de se considerar as incertezas da demanda no planejamento de redes de distribuição comparadas com modelos determinísticos tradicionais. O algoritmo proposto é testado em sistemas conhecidos na literatura. / This work proposes a methodology to solve distribution networks planning problems considering demand uncertainties. The methodology is based on the Artificial Immune System metaheuristics. In the solution process, the interval mathematics is implemented within the power flow and inside it, load uncertainties are modeled. The input interval variables are the active and reactive loads in the system buses. The expected result is presented in an interval and consists of minimizing the cable costs installation, construction or reinforcement of substations plus the costs associated with energy losses and operation considering a planning horizon. For this, restrictions such as radiality, connectivity, power balance and physical equipment capacities were used. To determine the best network topology, another methodology was adopted to compare the intervals. The main objective of this work is to evaluate the impact of considering the uncertainties of demand in the planning of distribution networks compared with traditional deterministic models. The proposed algorithm is tested in systems well-known in the literature.

Submitted by Geandra Rodrigues (geandrar@gmail.com) on 2018-10-16T12:46:06Z No. of bitstreams: 1 danielfioresibotelho.pdf: 2805489 bytes, checksum: 3d0b4a60f2a2f2cb7d26ea836254595a (MD5) / Approved for entry into archive by Adriana Oliveira (adriana.oliveira@ufjf.edu.br) on 2018-10-16T14:35:42Z (GMT) No. of bitstreams: 1 danielfioresibotelho.pdf: 2805489 bytes, checksum: 3d0b4a60f2a2f2cb7d26ea836254595a (MD5) / Made available in DSpace on 2018-10-16T14:35:42Z (GMT). No. of bitstreams: 1 danielfioresibotelho.pdf: 2805489 bytes, checksum: 3d0b4a60f2a2f2cb7d26ea836254595a (MD5) Previous issue date: 2018-08-31 / O presente trabalho apresenta uma metodologia para a alocação de unidades de Geração Distribuída no Sistema de Distribuição de Energia Elétrica visando a melhoria dos indicadores de confiabilidade da rede. O objetivo é avaliar qual o melhor ponto do sistema para realizar a inserção da unidade GD. Realiza-se também a análise de como essa alocação afeta os indicadores de qualidade e as perdas elétricas do sistema em estudo. A inclusão de limite para um indicador de qualidade considerado na distribuição de energia elétrica é explorada neste trabalho, verificando que a inserção de geração distribuída na rede pode fazer com que tal restrição seja atendida. A metodologia proposta é baseada nas técnicas meta-heurísticas Sistema Imunológico Artificial e Algoritmo Genético afim de comparar os resultados obtidos por ambas as técnicas. A função objetivo avaliada é composta por um somatório dos custos de confiabilidade, custos de investimento em geração distribuída e os custos das perdas elétricas do sistema; buscando assim sempre minimizar estes custos. Os algoritmos propostos são testados em sistemas conhecidos da literatura. / The present work presents a methodology for the allocation of Distributed Generation units in the Electric Energy Distribution System aiming at the improvement of network reliability indicators. The objective is to evaluate the best point of the system to perform the insertion of the GD unit. An analysis of how this allocation affects the quality indicators and the electrical losses of the system under study is also carried out. The inclusion of a limit for a quality indicator considered in the distribution of electric energy is explored in this work, verifying that the insertion of distributed generation in the network can cause that such restriction is met. The proposed methodology is based on the metaheuristic techniques Artificial Immune System and Genetic Algorithm in order to compare the results obtained by both techniques. The objective function evaluated is composed of a sum of the costs of reliability, costs of investment in distributed generation and the costs of the electrical losses of the system; always seeking to minimize these costs. The proposed algorithms are tested in systems known in the literature.

The main purpose of the natural immune system is to protect the body against any unwanted foreign cells that could infect the body and lead to devastating results. The nature immune system has different lymphocytes to detect and destroy these unwanted foreign patterns. The natural immune system can be modeled into an artificial immune system that can be used to detect any unwanted patterns in a non-biological environment. One of the main tasks of an immune system is to learn the structure of these unwanted patterns for a faster response to future foreign patterns with the same or similar structure. The artificial immune system (AIS) can therefore be seen as a pattern recognition system. The AIS contains artificial lymphocytes (ALC) that classify any pattern either as part of a predetermined set of patterns or not. In the immune system, lymphocytes have different states: Immature, Mature, Memory or Annihilated. Lymphocytes in the annihilated state needs to be removed from the active set of ALCs. The process of moving from one state to the next needs to be controlled in an efficient manner. This dissertation presents an AIS for detection of unwanted patterns with a dynamical active set of ALCs and proposes a threshold function to determine the state of an ALC. The AIS in the dissertation uses evolutionary computation techniques to evolve an optimal set of lymphocytes for better detection of unwanted patterns and removes ALCs in the annihilated state from the active set of ALCs. / Dissertation (MSc (Computer Science))--University of Pretoria, 2007. / Computer Science / unrestricted

Submitted by Renata Lopes (renatasil82@gmail.com) on 2016-01-19T10:47:34Z No. of bitstreams: 1 felipedasilvaseta.pdf: 1053075 bytes, checksum: 8a24a576cad55e9b46efe4bde9405104 (MD5) / Approved for entry into archive by Adriana Oliveira (adriana.oliveira@ufjf.edu.br) on 2016-01-25T17:44:56Z (GMT) No. of bitstreams: 1 felipedasilvaseta.pdf: 1053075 bytes, checksum: 8a24a576cad55e9b46efe4bde9405104 (MD5) / Made available in DSpace on 2016-01-25T17:44:56Z (GMT). No. of bitstreams: 1 felipedasilvaseta.pdf: 1053075 bytes, checksum: 8a24a576cad55e9b46efe4bde9405104 (MD5) Previous issue date: 2015-08-10 / CAPES - Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / O presente trabalho propõe uma metodologia para a resolução do problema de reconfiguração ótima de sistemas de distribuição de energia elétrica utilizando uma representação mais realista de parâmetros com incertezas. O objetivo é avaliar o impacto de se representar incertezas dos sistemas no problema de reconfiguração em relação a modelos tradicionais determinísticos. O modelo de reconfiguração probabilística proposto visa minimizar as perdas totais de energia considerando incertezas sobre a demanda e sobre a geração distribuída a partir da energia eólica, além de diferentes níveis de carregamento dos sistemas. A metodologia proposta é baseada na técnica meta-heurística Sistema Imunológico Artificial. Os fundamentos da matemática intervalar são incorporados em um fluxo de potência intervalar que modela as incertezas da demanda provenientes principalmente de erros de previsão e medição, bem como incertezas na geração por fontes eólicas devido a intermitências nos regimes de ventos. Desta forma, as variáveis de entrada intervalares são as demandas ativas e reativas das barras do sistema e os valores de velocidade de vento nas regiões das usinas eólicas. As incertezas da entrada são propagadas para as variáveis de saída do fluxo de potência, como as tensões nodais. Como resultado, as perdas totais de energia a serem minimizadas também são determinadas na forma intervalar. Uma metodologia para comparação de intervalos baseada na média e no raio dos intervalos é utilizada para determinar a topologia ótima. Restrições de tensão, radialidade e conectividade da rede são consideradas. O algoritmo proposto é testado em sistemas conhecidos da literatura. / The present work proposes a methodology to solve the problem of optimal reconfiguration of power distribution systems by using a more realistic representation of uncertain parameters. The objective is to evaluate the impact of representing uncertainties in the reconfiguration problem in relation to traditional deterministic models. The proposed probabilistic reconfiguration model aims at minimizing the total energy loss considering uncertainties on the load demand and the distributed generation from wind energy, as well as different load levels. The proposed methodology is based on the meta-heuristic technique Artificial Immune System. The interval mathematics fundamentals are embedded in an interval power flow that models the uncertainties of load forecast and measurements, as well as uncertainties due to the intermittences of the wind. Therefore, the input interval variables are the active and reactive loads at the network nodes and the wind speed in the regions where the wind farms are installed. The input uncertainties are thus propagated to the output power flow variables as the nodal voltages. As a result, the total energy losses to be minimized are also given in interval form. A methodology for comparing intervals that is based on the interval average and size is used to determine the best topology. Voltage constraints, radial configuration and network connectivity are considered. The proposed algorithm is tested in systems known in the literature.

Em anos recentes, muitos algoritmos bio-inspirados têm surgido para resolver problemas de classificação. Em confirmação a isso, a revista Nature, em 2002, publicou um artigo que já apontava para o ano de 2003 o uso comercial de Sistemas Imunológicos Artificiais para detecção de fraude em instituições financeiras por uma empresa britânica. Apesar disso, não observamos, a luz de nosso conhecimento, nenhuma publicação científica com resultados promissores desde então. Nosso trabalho tratou de aplicar Sistemas Imunológicos Artificiais (AIS) para detecção de fraude em cartões de crédito. Comparamos AIS com os métodos de Árvore de Decisão (DT), Redes Neurais (NN), Redes Bayesianas (BN) e Naive Bayes (NB). Para uma comparação mais justa entre os métodos, busca exaustiva e algoritmo genético (GA) foram utilizados para selecionar um conjunto paramétrico otimizado, no sentido de minimizar o custo de fraude na base de dados de cartões de crédito cedida por um emissor de cartões de crédito brasileiro. Em adição à essa otimização, fizemos também uma análise e busca por parâmetros mais robustos via multi-resolução, estes parâmetros são apresentados neste trabalho. Especificidades de bases de fraude como desbalanceamento de dados e o diferente custo entre falso positivo e negativo foram levadas em conta. Todas as execuções foram realizadas no Weka, um software público e Open Source, e sempre foram utilizadas bases de teste para validação dos classificadores. Os resultados obtidos são consistentes com Maes et al. que mostra que BN são melhores que NN e, embora NN seja um dos métodos mais utilizados hoje, para nossa base de dados e nossas implementações, encontra-se entre os piores métodos. Apesar do resultado pobre usando parâmetros default, AIS obteve o melhor resultado com os parâmetros otimizados pelo GA, o que levou DT e AIS a apresentarem os melhores e mais robustos resultados entre todos os métodos testados. / In 2002, January the 31st, the famous journal Nature, with a strong impact in the scientific environment, published some news about immune based systems. Among the different considered applications, we can find detection of fraudulent financial transactions. One can find there the possibility of a commercial use of such system as close as 2003, in a British company. In spite of that, we do not know of any scientific publication that uses Artificial Immune Systems in financial fraud detection. This work reports results very satisfactory on the application of Artificial Immune Systems (AIS) to credit card fraud detection. In fact, scientific financial fraud detection publications are quite rare, as point out Phua et al. [PLSG05], in particular for credit card transactions. Phua et al. points out the fact that no public database of financial fraud transactions is available for public tests as the main cause of such a small number of publications. Two of the most important publications in this subject that report results about their implementations are the prized Maes (2000), that compares Neural Networks and Bayesian Networks in credit card fraud detection, with a favored result for Bayesian Networks and Stolfo et al. (1997), that proposed the method AdaCost. This thesis joins both these works and publishes results in credit card fraud detection. Moreover, in spite the non availability of Maes data and implementations, we reproduce the results of their and amplify the set of comparisons in such a way to compare the methods Neural Networks, Bayesian Networks, and also Artificial Immune Systems, Decision Trees, and even the simple Naïve Bayes. We reproduce in certain way the results of Stolfo et al. (1997) when we verify that the usage of a cost sensitive meta-heuristics, in fact generalized from the generalization done from the AdaBoost to the AdaCost, applied to several tested methods substantially improves it performance for all methods, but Naive Bayes. Our analysis took into account the skewed nature of the dataset, as well as the need of a parametric adjustment, sometimes through the usage of genetic algorithms, in order to obtain the best results from each compared method.

La transformation de modèles consiste à transformer un modèle source en un modèle cible conformément à des méta-modèles source et cible. Nous distinguons deux types de transformations. La première est exogène où les méta-modèles source et cible représentent des formalismes différents et où tous les éléments du modèle source sont transformés. Quand elle concerne un même formalisme, la transformation est endogène. Ce type de transformation nécessite généralement deux étapes : l’identification des éléments du modèle source à transformer, puis la transformation de ces éléments. Dans le cadre de cette thèse, nous proposons trois principales contributions liées à ces problèmes de transformation. La première contribution est l’automatisation des transformations des modèles. Nous proposons de considérer le problème de transformation comme un problème d'optimisation combinatoire où un modèle cible peut être automatiquement généré à partir d'un nombre réduit d'exemples de transformations. Cette première contribution peut être appliquée aux transformations exogènes ou endogènes (après la détection des éléments à transformer). La deuxième contribution est liée à la transformation endogène où les éléments à transformer du modèle source doivent être détectés. Nous proposons une approche pour la détection des défauts de conception comme étape préalable au refactoring. Cette approche est inspirée du principe de la détection des virus par le système immunitaire humain, appelée sélection négative. L’idée consiste à utiliser de bonnes pratiques d’implémentation pour détecter les parties du code à risque. La troisième contribution vise à tester un mécanisme de transformation en utilisant une fonction oracle pour détecter les erreurs. Nous avons adapté le mécanisme de sélection négative qui consiste à considérer comme une erreur toute déviation entre les traces de transformation à évaluer et une base d’exemples contenant des traces de transformation de bonne qualité. La fonction oracle calcule cette dissimilarité et les erreurs sont ordonnées selon ce score. Les différentes contributions ont été évaluées sur d’importants projets et les résultats obtenus montrent leurs efficacités. / Model transformations take as input a source model and generate as output a target model. The source and target models conform to given meta-models. We distinguish between two transformation categories. Exogenous transformations are transformations between models expressed using different languages, and the whole source model is transformed. Endogenous transformations are transformations between models expressed in the same language. For endogenous transformations, two steps are needed: identifying the source model elements to transform and then applying the transformation on them. In this thesis, we propose three principal contributions. The first contribution aims to automate model transformations. The process is seen as an optimization problem where different transformation possibilities are evaluated and, for each possibility, a quality is associated depending on its conformity with a reference set of examples. This first contribution can be applied to exogenous as well as endogenous transformation (after determining the source model elements to transform). The second contribution is related precisely to the detection of elements concerned with endogenous transformations. In this context, we present a new technique for design defect detection. The detection is based on the notion that the more a code deviates from good practice, the more likely it is bad. Taking inspiration from artificial immune systems, we generate a set of detectors that characterize the ways in which a code can diverge from good practices. We then use these detectors to determine how far the code in the assessed systems deviates from normality. The third contribution concerns transformation mechanism testing. The proposed oracle function compares target test cases with a base of examples containing good quality transformation traces, and assigns a risk level based on the dissimilarity between the two. The traces help the tester understand the origin of an error. The three contributions are evaluated with real software projects and the obtained results confirm their efficiencies.

Submitted by Maicon Juliano Schmidt (maicons) on 2015-03-30T11:51:21Z No. of bitstreams: 1 Carlos Eduardo de Jesus Guimarães Oliveira.pdf: 1236896 bytes, checksum: ef9d04e6f25aee7908b56a622411bc74 (MD5) / Made available in DSpace on 2015-03-30T11:51:21Z (GMT). No. of bitstreams: 1 Carlos Eduardo de Jesus Guimarães Oliveira.pdf: 1236896 bytes, checksum: ef9d04e6f25aee7908b56a622411bc74 (MD5) Previous issue date: 2014-01-31 / Nenhuma / Este trabalho apresenta um algoritmo de otimização baseado na metaheurística dos Sistemas Imunológicos Artificiais, princípios de Teoria dos Jogos, Co-evolução e Paralelização. Busca-se a combinação adequada dos conceitos de Teoria dos Jogos, Co-evolução e Paralelização aplicados ao algoritmo AISO (Artificial Immune System Optimization) para resolução do Problema de Alocação de Berços (PAB). Dessa maneira, o algoritmo é formalizado a partir das técnicas citadas, formando o PCAISO-GT: Parallel Coevolutionary Artificial Immune System Optimization with Game Theory. Inicialmente, foram realizados experimentos visando à sintonia dos parâmetros empregados nas diferentes versões da ferramenta desenvolvida. Com base nas melhores configurações identificadas, foram realizados experimentos de avaliação através da solução de um conjunto de instâncias do PAB. Os resultados obtidos permitiram a indicação da versão co-evolutiva associada à teoria dos jogos como a melhor para solução do problema em estudo. / This paper presents an optimization algorithm based on metaheuristic of Artificial Immune Systems, principles of Game Theory, Co-evolution and parallelization. The objective is find the appropriate combination of the concepts of Game Theory, Co-evolution and Parallelization applied to AISO algorithm (Artificial Immune System Optimization) for solving the Berth Allocation Problem (BAP). Thus, the algorithm is formalized from the above mentioned techniques, forming the PCAISO-GT: Parallel Coevolutionary Artificial Immune System Optimization with Game Theory. Initially, experiments aiming to tune the parameters were performed using different versions of the tool developed. Based on the identified best settings, evaluation experiments were carried out by solving a set of instances of the PAB. The results obtained allowed the appointment of co-evolutionary version associated with game theory as the best solution to the problem under study.

La transformation de modèles consiste à transformer un modèle source en un modèle cible conformément à des méta-modèles source et cible. Nous distinguons deux types de transformations. La première est exogène où les méta-modèles source et cible représentent des formalismes différents et où tous les éléments du modèle source sont transformés. Quand elle concerne un même formalisme, la transformation est endogène. Ce type de transformation nécessite généralement deux étapes : l’identification des éléments du modèle source à transformer, puis la transformation de ces éléments. Dans le cadre de cette thèse, nous proposons trois principales contributions liées à ces problèmes de transformation. La première contribution est l’automatisation des transformations des modèles. Nous proposons de considérer le problème de transformation comme un problème d'optimisation combinatoire où un modèle cible peut être automatiquement généré à partir d'un nombre réduit d'exemples de transformations. Cette première contribution peut être appliquée aux transformations exogènes ou endogènes (après la détection des éléments à transformer). La deuxième contribution est liée à la transformation endogène où les éléments à transformer du modèle source doivent être détectés. Nous proposons une approche pour la détection des défauts de conception comme étape préalable au refactoring. Cette approche est inspirée du principe de la détection des virus par le système immunitaire humain, appelée sélection négative. L’idée consiste à utiliser de bonnes pratiques d’implémentation pour détecter les parties du code à risque. La troisième contribution vise à tester un mécanisme de transformation en utilisant une fonction oracle pour détecter les erreurs. Nous avons adapté le mécanisme de sélection négative qui consiste à considérer comme une erreur toute déviation entre les traces de transformation à évaluer et une base d’exemples contenant des traces de transformation de bonne qualité. La fonction oracle calcule cette dissimilarité et les erreurs sont ordonnées selon ce score. Les différentes contributions ont été évaluées sur d’importants projets et les résultats obtenus montrent leurs efficacités. / Model transformations take as input a source model and generate as output a target model. The source and target models conform to given meta-models. We distinguish between two transformation categories. Exogenous transformations are transformations between models expressed using different languages, and the whole source model is transformed. Endogenous transformations are transformations between models expressed in the same language. For endogenous transformations, two steps are needed: identifying the source model elements to transform and then applying the transformation on them. In this thesis, we propose three principal contributions. The first contribution aims to automate model transformations. The process is seen as an optimization problem where different transformation possibilities are evaluated and, for each possibility, a quality is associated depending on its conformity with a reference set of examples. This first contribution can be applied to exogenous as well as endogenous transformation (after determining the source model elements to transform). The second contribution is related precisely to the detection of elements concerned with endogenous transformations. In this context, we present a new technique for design defect detection. The detection is based on the notion that the more a code deviates from good practice, the more likely it is bad. Taking inspiration from artificial immune systems, we generate a set of detectors that characterize the ways in which a code can diverge from good practices. We then use these detectors to determine how far the code in the assessed systems deviates from normality. The third contribution concerns transformation mechanism testing. The proposed oracle function compares target test cases with a base of examples containing good quality transformation traces, and assigns a risk level based on the dissimilarity between the two. The traces help the tester understand the origin of an error. The three contributions are evaluated with real software projects and the obtained results confirm their efficiencies.