Les données industrielles offrent un matériau pour la prise de décision. Les travaux présentés concernent la transformation de données brutes en connaissances, pour contribuer au système de connaissances d’une organisation et améliorer son système décisionnel. Un processus d’aide à la décision est proposé. Il implique les acteurs de l’organisation et l’emploi de méthodes formelles. D’abord, il analyse et formalise les problématiques décisionnelles. Ensuite, il construit une aide la décision quantitative. Cette méthodologie est appliquée à un problème particulier : la gestion des capacités des TI d’une usine de STMicroelectronics. En effet, les managers doivent assurer un équilibre entre le coût de l’infrastructure TI et le niveau de service offert. Notre processus offre une aide pertinente. Il permet de surmonter deux enjeux, fréquents lors de la gestion des capacités : la complexité des systèmes IT et la prise en compte de l’activité métier. Situant ces travaux dans le cadre du référentiel ITIL, l’application du processus permet de constituer des modèles prédictifs, mettant en relation l’activité des serveurs informatiques et l’activité industrielle. Cette application permet aussi de contrôler dynamiquement la validité des modèles, ainsi que l’activité quotidienne du SI. Nos travaux formalisent quantitativement des connaissances, en favorisent l’utilisation dans les processus décisionnels, et en assurent l’évolution dans le temps. Nos recherches posent des fondations pour un plus large recours plus à l’exploitation des données issues des systèmes de production, dans le cadre du développement de systèmes de support à la décision et de perspectives Big Data. / A proper analysis of industrial data can provide material for decision making. The research work presented deals with the question: how can one convert raw data into useable information, to contribute to the knowledge management of an organization and improve its dynamic decision making? A decision-aid process is proposed. It implies actors of the organization and use of formal methods. Firstly, it analyses and formalizes decisional problems. Then, it develops an appropriate decision-aid on the basis of statistics. Our methodology is applied to a specific issue: capacity management of IT of a STMicroelectronics plant. This application raises a decision issue: managers have to ensure the right balance between infrastructure cost and service level offered. We demonstrate that the process may provide relevant support. This negates two managerial dilemmas, usually encountered when managing capacity: complexity of IT systems and incorporation of the business activity. Our application has been developed in the scope of the ITIL framework. It will be shown how the process builds predictive models, which link the activity of hardware servers to the industrial activity. Methods are also proposed, to monitor daily the quality of these models, as well as the overall activity of the IS. This work helps at formalizing quantitatively organizational knowledge, facilitating its use in decisional processes, but also ensuring its positive change over time. We hope this research is laying some foundations for a broader exploitation of the data stored in modern manufacturing systems, through future development of decision-support systems and Big Data initiatives.

The complex demographic changes that underlie the expansion of anatomically modern humans out of Africa have important consequences on the dynamics of natural selection and our ability to detect it. In this thesis, I aimed to refine our knowledge on human population history using ancient genomes, and then used a climate-informed, spatially explicit framework to explore the interplay between complex demographies and selection. I first analysed a high-coverage genome from Upper Palaeolithic Romania from ~37.8 kya, and demonstrated an early diversification of multiple lineages shortly after the out-of-Africa expansion (Chapter 2). I then investigated Late Upper Palaeolithic (~13.3ky old) and Mesolithic (~9.7 ky old) samples from the Caucasus and a Late Upper Palaeolithic (~13.7ky old) sample from Western Europe, and found that these two groups belong to distinct lineages that also diverged shortly after the out of Africa, ~45-60 ky ago (Chapter 3). Finally, I used East Asian samples from ~7.7ky ago to show that there has been a greater degree of genetic continuity in this region compared to Europe (Chapter 4). In the second part of my thesis, I used a climate-informed, spatially explicit demographic model that captures the out-of-Africa expansion to explore natural selection. I first investigated whether the model can represent the confounding effect of demography on selection statistics, when applied to neutral part of the genome (Chapter 5). Whilst the overlap between different selection statistics was somewhat underestimated by the model, the relationship between signals from different populations is generally well-captured. I then modelled natural selection in the same framework and investigated the spatial distribution of two genetic variants associated with a protective effect against malaria, sickle-cell anaemia and β⁰ thalassemia (Chapter 6). I found that although this model can reproduce the disjoint ranges of different variants typical of the former, it is incompatible with overlapping distributions characteristic of the latter. Furthermore, our model is compatible with the inferred single origin of sickle-cell disease in most regions, but it can not reproduce the presence of this disorder in India without long-distance migrations.