Les données hospitalières présentent de nombreuses spécificités qui rendent difficilement applicables les méthodes de fouille de données traditionnelles. Dans cette thèse, nous nous intéressons à l'hétérogénéité de ces données ainsi qu'à leur aspect temporel. Dans le cadre du projet ANR ClinMine et d'une convention CIFRE avec la société Alicante, nous proposons deux nouvelles méthodes d'extraction de connaissances adaptées à ces types de données. Dans la première partie, nous développons l'algorithme MOSC (Multi-Objective Sequence Classification) pour la classification supervisée sur données hétérogènes, numériques et temporelles. Cette méthode accepte, en plus des termes binaires ou symboliques, des termes numériques et des séquences d'événements temporels pour former des ensembles de règles de classification. MOSC est le premier algorithme de classification supportant simultanément ces types de données. Dans la seconde partie, nous proposons une méthode de biclustering pour données hétérogènes, un problème qui n'a à notre connaissance jamais été exploré. Cette méthode, HBC (Heterogeneous BiClustering), est étendue pour supporter les données temporelles de différents types : événements temporels et séries temporelles irrégulières. HBC est utilisée pour un cas d'étude sur un ensemble de données hospitalières, dont l'objectif est d'identifier des groupes de patients ayant des profils similaires. Les résultats obtenus sont cohérents et intéressants d'un point de vue médical ; et amènent à la définition de cas d'étude plus précis. L'intégration dans une solution logicielle est également engagée, avec une version parallèle de HBC et un outil de visualisation des résultats. / Hospital data exhibit numerous specificities that make the traditional data mining tools hard to apply. In this thesis, we focus on the heterogeneity associated with hospital data and on their temporal aspect. This work is done within the frame of the ANR ClinMine research project and a CIFRE partnership with the Alicante company. In this thesis, we propose two new knowledge discovery methods suited for hospital data, each able to perform a variety of tasks: classification, prediction, discovering patients profiles, etc.In the first part, we introduce MOSC (Multi-Objective Sequence Classification), an algorithm for supervised classification on heterogeneous, numeric and temporal data. In addition to binary and symbolic terms, this method uses numeric terms and sequences of temporal events to form sets of classification rules. MOSC is the first classification algorithm able to handle these types of data simultaneously. In the second part, we introduce HBC (Heterogeneous BiClustering), a biclustering algorithm for heterogeneous data, a problem that has never been studied so far. This algorithm is extended to support temporal data of various types: temporal events and unevenly-sampled time series. HBC is used for a case study on a set of hospital data, whose goal is to identify groups of patients sharing a similar profile. The results make sense from a medical viewpoint; they indicate that relevant, and sometimes new knowledge is extracted from the data. These results also lead to further, more precise case studies. The integration of HBC within a software is also engaged, with the implementation of a parallel version and a visualization tool for biclustering results.

An electron-positron linear collider is an option for future large particle accelerator projects. Such a collider would focus on precision tests of the Higgs boson properties. This thesis describes three studies related to the optimisation of highly granular calorimeters and one study on the sensitivity of Higgs couplings at CLIC. Photon reconstruction algorithms were developed for highly granular calorimeters of a future linear collider detector. A sophisticated pattern recognition algorithm was implemented, which uses the topological properties of electromagnetic showers to identify photon candidates and separate them from nearby particles. It performs clustering of the energy deposits in the detector, followed by topological characterisation of the clusters, with the results being considered by a multivariate likelihood analysis. This algorithm leads to a significant improvement in the reconstruction of both single photons and multiple photons in high energy jets compared to previous reconstruction software. The reconstruction and classification of tau lepton decay products was studied. Utilising highly granular calorimeters, the high resolution of energy and invariant mass of the tau decay products enabled a high classification rate. A hypothesis test was performed for expected decay final states. A multivariate analysis was trained to classify decay final states with a machine learning method. The performance of tau decay classification is used for the electromagnetic calorimeter optimisation at the ILC or CLIC. A proof-of-principle analysis using the correlation between the polarisations of the tau pair from a boson decay as a signature to differentiate the Higgs boson from the Z boson is presented. Sensitivity of Higgs couplings at CLIC was studied using the double Higgs production process. Algorithms were developed for signal event selection. The event selection relies on the jet reconstruction and the flavour tagging. A multivariate analysis is performed to select signal events. An attempt at extracting Higgs trilinear self-coupling and quartic coupling was conducted.