Motivated by IT evaluation problems identified in a large public sector organization, we propose how evaluation requirements can be supported by a framework combining different models and methods from IS evaluation theory. The article extends the content, context, process (CCP) perspectives of organizational change with operations research techniques and demonstrates the approach in practice for an Enterprise Resource Planning evaluation.

Os avanços na Tecnologia da Informação e a proliferação de itens de consumo, entre outros aspectos, mudaram o cenário e o desempenho das previsões. Os processos de previsão devem ser reexaminados, estabelecendo mecanismos de comunicação formais que compartilhem a informação entre os diferentes níveis hierárquicos dentro da organização, eliminando ou reduzindo o desconforto das previsões paralelas e desconexas oriundas de níveis hierárquicos diferentes. O objetivo deste trabalho é propor um sistema de apoio à decisão baseado em métodos matemáticos e sistemas de informação, capaz de integrar as previsões de vários níveis hierárquicos de uma empresa por um repositório de dados (Data Warehouse ou DW) e um Sistema de Apoio à Decisão (SAD) com sistema Business Intelligence (BI), onde os níveis hierárquicos acessem as informações com o nível de detalhe apropriado dentro do processo de decisão, alinhado às expectativas corporativas de crescimento. Assim, a modelagem realizada neste trabalho teve como foco a geração de cenários para criar um sistema de apoio à decisão, prevendo demandas agregadas e individuais, gerando uma estrutura de integração entre as previsões feitas em diferentes níveis e alinhando valores oriundos de métodos quantitativos e julgamento humano. Uma das maiores preocupações foi verificar qual método (séries temporais, métodos causais) teria destaque em um processo integrado de previsão. Entre os diferentes testes efetuados, pode-se destacar os seguintes resultados: (1) a suavização exponencial tripla proporcionou melhor ajuste (dos dados passados) de séries históricas de demandas mais agregadas e proporcionou previsões mais precisas de representatividades agregadas. Para séries históricas de demanda individual e representatividade individual, os outros métodos comparados apresentaram desempenho muito próximo; (2) a criação de diferentes cenários de previsão, fazendo uso de um repositório de dados e sistema de apoio à decisão, permitiu análise de uma gama de diferentes valores futuros. Uma forma de simulação para apoiar a formulação das expectativas da diretoria foi adaptada da literatura e sugerida; (3) os erros de previsão nas abordagens top-down ou bottom-up são estatisticamente iguais no contexto desta pesquisa. Conclui-se que o método de suavização exponencial tripla traz menos erros às previsões de séries mais agregadas, se comparado com outros métodos abordados no trabalho. Esse fato está de acordo com asserções encontradas na literatura pesquisada de que o método de suavização exponencial é cada vez mais utilizado na previsão, em detrimento dos métodos causais como a regressão múltipla. Conclui-se, principalmente, que os sistemas SAD e BI propostos deram suporte aos vários níveis hierárquicos, proporcionando variedades de estilos de decisão e que podem diminuir o hiato entre o raciocínio qualitativo adotado em nível estratégico e o aspecto quantitativo mais comum em níveis operacionais em qualquer empresa. / Advances in Information Technology (IT), and the increase of consumption items, among other things, changed the performance in the forecasts predictions. It is not uncommon that organizations will perform parallel forecasts within the various hierarchical levels without communicating with each other. The objective of this work is to build an integrated \"infrastructure\" for forecasting through a repository of data (Data Warehouse or DW) and a Decision Support System (DSS) with Business Intelligence (BI) where the hierarchical levels have access to the information with the appropriate level of detail within the process, aligned to the corporate growth expectations. The modeling in this work focused in the generation of scenarios to create a decision support system, predicting individual and aggregate demand, create a structure for integrating and aligning the estimated forecast generated by quantitative and qualitative methods. After a series of experimental tests, main results found were: (1) triple exponential smoothing provided the best fit using historical aggregated demand, and provided a more precise estimate of aggregate representation. For historical series of individual demand and individual representation, the other methods used for comparison performed similarly; (2) the creation of different scenarios for prediction, using data repository and decision support system, allowed for analysis of a range of different future values. The simulation to support management expectations has been adapted from the literature; (3) the prediction errors in the top-down and bottom-up approaches are statistically the same in the context of this research. In conclusion, the method of triple exponential smoothing has fewer errors in the forecasts of aggregated series when compared to other methods discussed in this work. Moreover, the DSS and BI systems provided decision-making support to the various hierarchical levels, reducing the gap between qualitative and quantitative decision processes thus bridging the strategic and operational decision making processes.

Este trabalho apresenta a construção de um sistema especialista aplicado ao diagnóstico de doenças cardíacas, usando como ferramenta computacional redes Bayesianas. O trabalho envolveu a interação entre diferentes áreas do conhecimento, engenharia e medicina, com maior foco na metodologia da construção de sistemas especialistas. São apresentados os processos de definição do problema, modelagem qualitativa e quantitativa, e avaliação. Neste trabalho, os processos de modelagem e avaliação foram realizados com o auxílio de um especialista médico e de dados bibliográficos. São apresentados como resultados a rede Bayesiana construída e um software para manipulação de redes Bayesianas denominado iBNetz. / This work presents the construction of an expert system applied to the diagnosis of heart diseases, using Bayesian networks as a modeling tool. The work involved interactions between two different fields, engineering and medicine, with special emphasis on the methodology of building expert systems. The processes of problem definition, qualitative and quantitative modeling, and evaluation are presented here. In this work, the modeling and evaluation processes have been conducted with the aid of a medical expert and bibliographic sources. The work has produced a Bayesian network for diagnosis and a software, called iBNetz, for creating and manipulating Bayesian networks.

L'objectif de cette thèse est de fournir des outils d'aide à la décision pour piloter les plateformes logistiques à court de moyen terme. La première partie décrit la problématique concernée et les notions essentielles dans le cadre des chaînes logistiques. Dans la deuxième partie, le problème de la planification est étudié, nous proposons des modèles linéaires pour minimiser les coûts de personnel, qui prennent en compte les flux : leurs variations saisonnières, la possibilité de les négocier localement en amont et en aval, ainsi que leur organisation et celle du travail. Ainsi, l'outil peut être utilisé dans la coordination des flux entres les partenaires de la chaîne livrées en amont et en aval de la plateforme et la négociation des dates de livraison. Ces modèles sont testés et validés sur des instances générées aléatoirement, sur des configurations inspirées de problèmes réels. Dans la troisième partie, nous travaillons sur l'ordonnancement des activités de préparation de commandes. Ici, nous combinons deux familles de contraintes difficiles : l'arrivée de composants (ressources consommables) à des dates et en quantités connues à l'amont de la plateforme, et des tournées de livraison à dates fixées à l'aval. Trois cas particuliers sont étudiés, selon la façon dont les tournées sont organisées. Nous proposons des procédures par séparation et évaluation pour ces problèmes, et un modèle linéaire en nombres entiers pour le cas le plus simple. Des expériences sont faites sur des familles d'instances générées aléatoirement et de manière partiellement hétérogène. Plusieurs perspectives de généralisation sont proposées / The aim of this thesis is to provide decision support systems to control logistic platforms at the mid-term and short-term levels. Several problems and main notions concerning logistic platform context are described in the first part. In the second part, planning problems are studied. Two linear programming models are proposed to minimize the workforce costs. These models take into account several characteristics : seasonal flow variations, work and flow organization in the platform, and local negotiations of the upstream and downstream flows. In consequence, our decision support system can be used in the flow coordination between supply chain partners. Two types of negotiations are considered : negotiations of upstream and downstream delivered quantities and negotiation of delivery dates. These models have been tested on pertinent randomly generated instances inspired from concerete problems. In the third part of the thesis, the external flows of the platforme are assumed to be fixed. Orders preparation scheduling problem inside the platform is considered. Two families of strong contraints are combined : staircase availability of components (consumable resources) and dixed delivery dates. According to the way the downstream deliveries are organized and penalised, three different cases (based on industrial applications) have been studied. We proposed three branch and bound procedures for these problems, and an integer linear program for the easiest problem. Experimental analysis has been done over heterogeneous randomly generated instance families. In the last part, a series of perspectives for this work are proposed

This dissertation discusses three problems from different areas of medical research and their machine learning solutions. Each solution is a distinct type of decision support system. They show three common properties: personalized healthcare decision support, reduction of the use of medical resources, and improvement of outcomes. The first decision support system assists individual hospital selection. This system can help a user make the best decision in terms of the combination of mortality, complication, and travel distance. Both machine learning and optimization techniques are utilized in this type of decision support system. Machine learning methods, such as Support Vector Machines, learn a decision function. Next, the function is transformed into an objective function and then optimization methods are used to find the values of decision variables to reach the desired outcome with the most confidence. The second decision support system assists diagnostic decisions in a sequential decision-making setting by finding the most promising tests and suggesting a diagnosis. The system can speed up the diagnostic process, reduce overuse of medical tests, save costs, and improve the accuracy of diagnosis. In this study, the system finds the test most likely to confirm a diagnosis based on the pre-test probability computed from the patient's information including symptoms and the results of previous tests. If the patient's disease post-test probability is higher than the treatment threshold, a diagnostic decision will be made, and vice versa. Otherwise, the patient needs more tests to help make a decision. The system will then recommend the next optimal test and repeat the same process. The third decision support system recommends the best lifestyle changes for an individual to lower the risk of cardiovascular disease (CVD). As in the hospital recommendation system, machine learning and optimization are combined to capture the relationship between lifestyle and CVD, and then generate recommendations based on individual factors including preference and physical condition. The results demonstrate several recommendation strategies: a whole plan of lifestyle changes, a package of n lifestyle changes, and the compensatory plan (the plan that compensates for unwanted lifestyle changes or real-world limitations).

Drug effects can be unpredictable and vary widely among patients with environmental, genetic, and clinical factors. Randomized control trials (RCTs) are not sufficient to identify adverse drug reactions (ADRs), and the electronic health record (EHR) along with medical claims have become an important resource for pharmacovigilance. Among all the data collected in hospitals, laboratory tests represent the most documented and reliable data type in the EHR. Laboratory tests are at the core of the clinical decision process and are used for diagnosis, monitoring, screening, and research by physicians. They can be linked to drug effects either directly, with therapeutic drug monitoring (TDM), or indirectly using drug laboratory effects (DLEs) that affect surrogate tests. Unfortunately, very few automated methods use laboratory tests to inform clinical decision making and predict drug effects, partly due to the complexity of these time series that are irregularly sampled, highly dependent on other clinical covariates, and non-stationary. Deep learning, the branch of machine learning that relies on high-capacity artificial neural networks, has known a renewed popularity this past decade and has transformed fields such as computer vision and natural language processing. Deep learning holds the promise of better performances compared to established machine learning models, although with the necessity for larger training datasets due to their higher degrees of freedom. These models are more flexible with multi-modal inputs and can make sense of large amounts of features without extensive engineering. Both qualities make deep learning models ideal candidate for complex, multi-modal, noisy healthcare datasets. With the development of novel deep learning methods such as generative adversarial networks (GANs), there is an unprecedented opportunity to learn how to augment existing clinical dataset with realistic synthetic data and increase predictive performances. Moreover, GANs have the potential to simulate effects of individual covariates such as drug exposures by leveraging the properties of implicit generative models. In this dissertation, I present a body of work that aims at paving the way for next generation laboratory test-based clinical decision support systems powered by deep learning. To this end, I organized my experiments around three building blocks: (1) the evaluation of various deep learning architectures with laboratory test time series and their covariates with a forecasting task; (2) the development of implicit generative models of laboratory test time series using the Wasserstein GAN framework; (3) the inference properties of these models for the simulation of drug effects in laboratory test time series, and their application for data augmentation. Each component has its own evaluation: The forecasting task enabled me to explore the properties and performances of different learning architectures; the Wasserstein GAN models are evaluated with both intrinsic metrics and extrinsic tasks, and I always set baselines to avoid providing results in a "neural-network only" referential. Applied machine learning, and more so with deep learning, is an empirical science. While the datasets used in this dissertation are not publicly available due to patient privacy regulation, I described pre-processing steps, hyper-parameters selection and training processes with reproducibility and transparency in mind. In the specific context of these studies involving laboratory test time series and their clinical covariates, I found that for supervised tasks, machine learning holds up well against deep learning methods. Complex recurrent architectures like long short-term memory (LSTM) do not perform well on these short time series, while convolutional neural networks (CNNs) and multi-layer perceptrons (MLPs) provide the best performances, at the cost of extensive hyper-parameter tuning. Generative adversarial networks, enabled by deep learning models, were able to generate high-fidelity laboratory test time series, and the quality of the generated samples was increased with conditional models using drug exposures as auxiliary information. Interestingly, forecasting models trained on synthetic data exclusively still retain good performances, confirming the potential of GANs in privacy-oriented applications. Finally, conditional GANs demonstrated an ability to interpolate samples from drug exposure combinations not seen during training, opening the way for laboratory test simulation with larger auxiliary information spaces. In specific cases, augmenting real training sets with synthetic data improved performances in the forecasting tasks, and could be extended to other applications where rare cases present a high prediction error.

Abstract not available

Abstract not available

Abstract not available

Existing patient medical records are a rich data source with a potential to support clinical research. Fragmentation of data across disparate medical database inhibits the use of these existing datasets. Overcoming such disjointedness is possible through the use of a data warehouse. Once the data is cleansed, transformed, and stored within the data warehouse it is possible to turn attention to the exploration of the medical datasets. Exploratory and confirmatory Data Mining Tools are well suited to such activities. This thesis concerned with: demonstrating parallels between scientific method and CRISP-DM; extending CRISP-DM for use with medical datasets; and proposal of the supporting Intelligent Decision Support System framework. This research has been undertaken using a fetal-maternal case study. / Master of Science (Hons)