No description available.

Time series is a collection of points gathered at regular intervals. Time series analysis explores the time correlations and tries to model it according to trend and seasonality. One of the most relevant tasks, in time series analysis, is forecasting future values, which is considered fundamental in many real-world scenarios. Nowadays, many companies forecast using hand-written models or naive statistical models. Call centers are the front end of the organization, managing the relationship with the customers. A key challenge for call centers remains the call load forecast and the optimization of the schedule. Call load indicates the number of calls a call center receives. The call load forecast is mostly exploited to schedule the staff. They are interested in the short term forecast to handle the unforeseen and to optimize the staff schedule, and in the long term forecast to hire or assign staff to other tasks. Machine learning has been applied to several fields reporting excellent results, and recently, time series forecasting problems have gained a high-interest thanks to the new recurrent network, named Long-short Term Memory. This thesis has explored the capabilities of machine learning in modeling and forecasting call load time series, characterized by a strong seasonality, both at daily and hourly scale. We compare Seasonal Artificial Neural Network (ANN) and a Long-Short Term Memory (LSTM) models with Seasonal Autoregressive Integrated Moving Average (SARIMA) model, which is one of the most common statistical method utilized by call centers. The primary metric used to evaluate the results is the Normalized Mean Squared Error (NMSE), the secondary is the Symmetric Mean Absolute Percentage Error (SMAPE), utilized to calculate the accuracy of the models. We carried out our experiments on three different datasets provided by the Teleopti. Experimental results have proven SARIMA to be more accurate in forecasting at daily scale across the three datasets. It performs better than the Seasonal ANN and the LSTM with a limited amount of data points. At hourly scale, Seasonal ANN and LSTM outperform SARIMA, showing robustness across a forecasting horizon of 160 points. Finally, SARIMA has shown no correlation between the quality of the model and the number of data points, while both SANN and LSTM improves together with the number of sample / Tidsserie är en samling punkter som samlas in med jämna mellanrum. Tidsseriens analys undersöker tidskorrelationerna och försöker modellera den enligt trend och säsongsbetonade. En av de mest relevanta uppgifterna, i tidsserieranalys, är att förutse framtida värden, som anses vara grundläggande i många verkliga scenarier. Numera förutspår många företag med handskrivna modeller eller naiva statistiska modeller. Callcenter är organisationens främre del och hanterar relationen med kunderna. En viktig utmaning för callcentra är fortfarande samtalslastprognosen och optimeringen av schemat. Samtalslast indikerar antalet samtal ett callcenter tar emot. Samtalslastprognosen utnyttjas mest för att schemalägga personalen. De är intresserade av den kortsiktiga prognosen för att hantera det oförutsedda och för att optimera personalplanen och på långsiktigt prognos för att anställa eller tilldela personal till andra uppgifter. Maskininlärning har använts på flera fält som rapporterar utmärkta resultat, och nyligen har prognosproblem i tidsserier fått ett stort intresse tack vare det nya återkommande nätverket, som heter Long-short Term Memory. Den här avhandlingen har undersökt kapaciteten för maskininlärning i modellering och prognoser samtalsbelastningstidsserier, kännetecknad av en stark säsongsbetonning, både på daglig och timskala. Vi jämför modeller med säsongsmässigt artificiellt neuralt nätverk (ANN) och ett LSTM-modell (Long- Short Term Memory) med Seasonal Autoregressive Integrated Moving Average (SARIMA)-modell, som är en av de vanligaste statistiska metoderna som används av callcenter. Den primära metriken som används för att utvärdera resultaten är det normaliserade medelkvadratfelet (NMSE), det sekundära är det symmetriska genomsnittet absolut procentuellt fel (SMAPE), som används för att beräkna modellernas noggrannhet. Vi genomförde våra experiment på tre olika datasätt från Teleopti. Experimentella resultat har visat att SARIMA är mer exakt när det gäller prognoser i daglig skala över de tre datasätten. Det presterar bättre än Seasonal ANN och LSTM med en begränsad mängd datapoäng. På timskala överträffar Seasonal ANN och LSTM SARIMA och visar robusthet över en prognoshorisont på 160 poäng. SARIMA har slutligen inte visat någon korrelation mellan modellens kvalitet och antalet datapunkter, medan både SANN och LSTM förbättras tillsammans med antalet sampel.

Le pneumocoque est une cause majeure d'infections bactériennes communautaires dans le monde. D'après la littérature, la consommation d'antibiotiques pourrait influer sur le risque de colonisation ou d'infection par pneumocoque à sensibilité diminuée aux antibiotiques spécifiques. La France, grande consommatrice d'antibiotiques, a mis en place, à l'automne 2002, un plan national pour préserver l'efficacité des antibiotiques et améliorer leur usage. Cette campagne a conduit à une diminution significative de la consommation d'antibiotiques durant les périodes hivernales. En 2003, une vaccination anti-pneumococcique des enfants de moins de deux ans a été recommandée afin de réduire les infections communautaires à pneumocoque chez l'enfant. Au vu du contexte français, il paraît important d'étudier la dynamique des infections invasives communautaires à pneumocoque en prenant en compte les deux interventions de santé publique. L'étude de l'association entre deux ou plusieurs séries temporelles saisonnières doit être effectuée sur des séries stationnarisées afin d'éliminer tout risque de confusion. Les différentes méthodes de désaisonnalisation ont été comparées à travers une étude de simulations afin d'identifier la stratégie optimale. De plus, le modèle de régression linéaire adapté aux séries temporelles repose sur l'hypothèse de la linéarité du lien. Cependant, cette hypothèse est critiquable en particulier lorsqu'on s'intéresse au lien associé à une série de type épidémique. Une deuxième étude de simulations a été réalisée afin d'étudier l'impact de l'hypothèse de la linéarité du lien lors de son estimation.A partir des stratégies permettant d'étudier le lien entre plusieurs séries saisonnières identifiées à partir des études de simulations, la dynamique des infections invasives communautaires à pneumocoque a été étudiée en France entre janvier 2002 et décembre 2009. / Streptococcus pneumoniae is a leading cause of communitary-acquired pneumococcal invasive infections worldwide. Recent surveys studied the association between pneumococcal carriage and antibiotic consumption. Reducing antibiotic consumption migth reduce pneumococcal carriage. In France, a national campaign was launched in 2002 in order to reduce antibiotic consumption mainly in the community. In 2003, the seven-valent pneumococcal conjugate vaccine was introduced and recommanded for to children in order to reduce the risk of invasive pneumococcal infections. In this contexte, it is worth investigating the evolution of communitary-acquired pneumococcal invasive infections in France.When examining the association between two monthly time series data with some common seasonal pattern, we are faced with the problem of eliminating this seasonal variation. Indeed this common seasonal feature will act as a confounder if not removed. Even if several methods exist, such as the use of semi-parametric or trigonometric functions, no optimal method has been yet identified. Hence, we compared performances of available smoothing approaches to estimate a temporal link between two series using extensive simulations. The linear regression usually used to estimate the link between two time series is based on the hypothesis of a linear link. However, such a link might not be linear when considering an association with an epidemic time series. In order to check whether this linear model can also manage non linear relationships, a simulation study was also settled. Finally, from these simulation studies, we identified strategies that where implemented to estimate the association between community-acquired pneumococcal invasive infections and antibiotic exposure.

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

Submitted by Rosivalda Pereira (mrs.pereira@ufma.br) on 2017-07-03T18:32:31Z No. of bitstreams: 1 SelmoRodrigues.pdf: 1374245 bytes, checksum: 96afcfa04ba5cc18c4db55e4c92cdf23 (MD5) / Made available in DSpace on 2017-07-03T18:32:31Z (GMT). No. of bitstreams: 1 SelmoRodrigues.pdf: 1374245 bytes, checksum: 96afcfa04ba5cc18c4db55e4c92cdf23 (MD5) Previous issue date: 2017-01-26 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) / This paper proposes a new methodology for modelling based on an evolving Neuro-Fuzzy Network Takagi-Sugeno (NFN-TS) for seasonal time series forecasting. The NFN-TS use the unobservable components extracted from the time series to evolve, i.e., to adapt and to adjust its structure, where the number of fuzzy rules of this network can increase or reduced according the components behavior. The method used to extract the components is a recursive version developed in this paper based on the Spectral Singular Analysis (SSA) technique. The proposed methodology has the principle divide to conquer, i.e., it divides a problem into easier subproblems, forecasting separately each component because they present dynamic behaviors that are simpler to forecast. The consequent propositions of fuzzy rules are linear state space models, where the states are the unobservable components data. When there are available observations from the time series, the training stage of NFN-TS is performed, i.e., the NFN-TS evolves its structure and adapts its parameters to carry out the mapping between the components data and the available sample of original time series. On the other hand, if this observation is not available, the network considers the forecasting stage, keeping its structure fixed and using the states of consequent fuzzy rules to feedback the components data to NFN-TS. The NFN-TS was evaluated and compared with other recent and traditional techniques for forecasting seasonal time series, obtaining competitive and advantageous results in relation to other papers. This paper also presents a case study of proposed methodology for real-time detection of anomalies based on a patient’s electrocardiogram data. / Esse trabalho propõe uma nova metodologia para modelagem baseada em uma Rede Neuro- Fuzzy Takagi-Sugeno (RNF-TS) evolutiva para a previsão de séries temporais sazonais. A RNF-TS considera as componentes não-observáveis extraídas a partir da série para evoluir, ou seja, adaptar e ajustar sua estrutura, sendo que a quantidade de regras fuzzy dessa rede pode aumentar ou ser reduzida conforme o comportamento das componentes. O método utilizado para extrair as componentes é uma versão recursiva desenvolvida nessa pesquisa baseada na técnica de Análise Espectral Singular (AES). A metodologia proposta tem como princípio dividir para conquistar, isto é, dividir um problema em subproblemas mais fáceis de lidar, realizando a previsão separadamente de cada componente já que apresentam comportamentos dinâmicos mais simples de prever. As proposições do consequente das regras fuzzy são modelos lineares no espaço de estados, sendo que os estados são os próprios dados das componentes não-observáveis. Quando há observações disponíveis da série temporal, o estágio de treinamento da RNF-TS é realizado, ou seja, a RNF-TS evolui sua estrutura e adapta seus parâmetros para realizar o mapeamento entre os dados das componentes e a amostra disponível da série temporal original. Caso contrário, se essa observação não está disponível, a rede aciona o estágio de previsão, mantendo sua estrutura fixa e usando os estados dos consequentes das regras fuzzy para realimentar os dados das componentes para a RNF-TS. A RNF-TS foi avaliada e comparada com outras técnicas recentes e tradicionais para previsão de séries temporais sazonais, obtendo resultados competitivos e vantajosos em relação a outras pesquisas. Este trabalho apresenta também um estudo de caso da metodologia proposta para detecção em tempo-real de anomalias baseada em dados de eletrocardiogramas de um paciente.

本論文主要研究以類神經網路模式預測季節性時間序列之有效性。利用適 當地建構樣本訓練集,網路經訓練後可作為季節性時間序列之預測工具。 文中亦提出移動學習法以期提高預測之準確度。並以台灣地區每季進口商 品與勞務總值則作為實證之研究。此季節性時間序列因受離群值之影響而 增加其預測困難度。實證結果顯示類神經網路模式之預測表現較傳統之統 計方法優異,即使此序列受到離群值之干擾。 / We investigate the effectiveness of neural networks for predicting the future behavior of seasonal time series. Utilizing the training set constructed properly, we can train the network who can be used to predict the future of seasonal time series. A shifting-learning method is also employed in order to obtained a better forecasting performance. The quarterly imports of goods and services of Taiwan between the first quarter of 1968 and the fourth quarter of 1990 are studied in the research. The series are contaminated with outliers, which will increase the difficulty of forecasting. Empirical results exhibit that neural networks model free approach have better prediction performance than the classical Box-Jenkins approach, even the series are contaminated with outliers.

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