Financial time-series forecasting has long been discussed because of its vitality for making informed investment decisions. This kind of problem, however, is intrinsically challenging due to the data dynamics in nature. Most of the research works in the past focus on artificial neural network (ANN)-based approaches. It has been pointed out that such approaches suffer from explanatory power and generalized prediction ability though. The objective of this research is thus to propose a hybrid approach for stock price forecasting. Independent component analysis (ICA) is employed to reveal the latent structure of the observed time-series and remove noise and redundancy in the structure. It further assists clustering analysis. Support vector regression (SVR) models are then applied to enhance the generalization ability with separate models built based on the time-series data of companies in each individual cluster. Two experiments are conducted accordingly. The results show that SVR has robust accuracy performance. More importantly, SVR models with ICA-based clustered data perform better than the single SVR model with all data involved. Our proposed approach does enhance the generalization ability of the forecasting models, which justifies the feasibility of its applications.

Emergency department (ED) overcrowding is a challenge faced by many hospitals. One approach to mitigate overcrowding is to anticipate high levels of overcrowding. The purpose of this study was to forecast a measure of ED overcrowding four hours in advance to allow clinicians to prepare for high levels of overcrowding. The chosen measure of ED overcrowding was ED length of stay compliance measures set by the Ontario government. A feed-forward artificial neural network (ANN) was designed to perform a time series forecast on the number of patients that were non-compliant. Using the ANN compared to historical averages, a 70% reduction in the root mean squared error was observed as well as good discriminatory ability of the ANN model with an area under the receiver operating characteristic curve of 0.804. Therefore, using ANNs to forecast ED overcrowding gives clinicians an opportunity to be proactive, rather than reactive, in ED overcrowding crises.

Emergency department (ED) overcrowding is a challenge faced by many hospitals. One approach to mitigate overcrowding is to anticipate high levels of overcrowding. The purpose of this study was to forecast a measure of ED overcrowding four hours in advance to allow clinicians to prepare for high levels of overcrowding. The chosen measure of ED overcrowding was ED length of stay compliance measures set by the Ontario government. A feed-forward artificial neural network (ANN) was designed to perform a time series forecast on the number of patients that were non-compliant. Using the ANN compared to historical averages, a 70% reduction in the root mean squared error was observed as well as good discriminatory ability of the ANN model with an area under the receiver operating characteristic curve of 0.804. Therefore, using ANNs to forecast ED overcrowding gives clinicians an opportunity to be proactive, rather than reactive, in ED overcrowding crises.

Στην εργασία που ακολουθεί μελετήθηκε η χρήση μεθόδων της υπολογιστικής νοημοσύνης στην πρόβλεψη της κίνησης της ισοτιμίας νομισμάτων. Για να γίνει αυτό αναπτύχθηκε ένας υβριδικός αλγόριθμος που χρησιμοποιεί νευρωνικά δύκτια και γενετικούς αλγόριθμους. Στο Πρώτο Κεφάλαιο παρουσιάζεται η ϑεωρία των νευρωνικών δικτύων, οι αρχιτεκτονικές και οι μέθοδοι εκπαίδευσής τους. Επιπλέον παρουσιάζονται οι γενετικοί αλγόριθμοι και ο γενικός τρόπος λειτουργίας τους. Στο Δεύτερο Κεφάλαιο εξετάζεται το πρόβλημα της πρόγνωσης, από την σκοπιά των νευρωνικών δικτύων, καθώς και η προβλεψιμότητα των οικονομικών χρονοσειρών. Επιπλέον παρουσιάζονται υβριδικά συστήματα που χρησιμοποιούνται για την πρόβλεψη χρονοσειρών και επεξηγείται ο τρόπος λειτουργίας του αλγορίθμου που αναπτύχθηκε εδώ. Επιπλέον παρατίθενται τα αποτελέσματα της χρήσης του λογισμικού που αναπτύχθηκε, στην πρόγνωση της μεταβολής της ισοτιμίας νομισμάτων. Στο Παράρτημα παρέχεται ο πλήρης κώδικας που αναπτύχθηκε σε MATLAB. / --

Η πρόβλεψη χρονοσειρών και μάλιστα αποτελούμενων από χρηματοοικονομικά δεδομένα έχει αποτελέσει αντικείμενο εκτεταμένης ερευνητικής δραστηριότητας. Στη χρηματοοικονομική επιστήμη, η ανάλυση χρονοσειρών εφαρμόζεται ευρέως για την πρόβλεψη των τιμών των διεθνών και εθνικών χρηματαγορών αλλά και σε εφαρμογές σχετικές με τη διαδικασία πρόβλεψης χρηματοοικονομικών κρίσεων. Η βασική διαφοροποίηση της διατριβής αυτής έγκειται στο αντικείμενο της πρόβλεψης· αντί της επικέντρωσης στην εύρεση της μελλοντικής τιμής μίας χρονοσειράς, οι παραγόμενες προβλέψεις στοχεύουν στον χρονικό εντοπισμό του μελλοντικού σημείου στο οποίο μία χρονοσειρά αναμένεται να βελτιστοποιηθεί τοπικά. Η παρούσα διατριβή πραγματεύεται την εισαγωγή μίας τεχνικής οπισθοδρόμησης η οποία εξομοιώνει διάφορες τεχνικές βελτιστοποίησης. Οι προτεινόμενες παραλλαγές της τεχνικής οπισθοδρόμησης καταλήγουν στη δημιουργία μεθοδολογιών οι οποίες στοχεύουν στην επίλυση προβλημάτων εντοπισμού του χρόνου παρουσίασης του τοπικού μελλοντικού βέλτιστου της εξεταζόμενης χρονοσειράς. Επιπλέον, η τεχνική προσφέρει και μεθοδολογικό πλαίσιο προς εξέταση του ζητήματος της ex ante πρόβλεψης μίας χρηματοοικονομικής κρίσης. Από την επενδυτική σκοπιά, οι πληροφορίες αυτές μπορεί να αποτελέσουν χρήσιμο εργαλείο υιοθέτησης επενδυτικής στρατηγικής και διαχείρισης χαρτοφυλακίου. Τέλος, η εμπειρική έρευνα καταλήγει στην εφαρμογή της προτεινόμενης τεχνικής σε δεδομένα από βασικές χρηματοπιστωτικές αγορές σε παγκόσμια κλίμακα αλλά και στην εγχώρια αγορά. / Time series prediction, especially in the case of financial time series, has attracted major research interest. In finance, time series analysis is applied widely for the purposes of predicting prices of international and national markets; also it is used for the prediction of financial crises. This thesis differences in the prediction’s objective; instead of focusing on the time series’ future price it aims on detecting the future time that the time series is expected to be locally optimized. This thesis introduces a backtracking techniques that integrates elements of specific optimization techniques. The introduced variations of the technique generate methodologies that confront the problem of the chronical allocation of a time series’ local optima. Moreover, the technique provides a methodological frame for the examination of the ex ante prediction of a financial crisis. Under the investment spectrum such information may provide a useful tool for the adoption of investment strategy and portfolio management. Finally the empirical research concludes with the application of the proposed techniques to data deriving from major financial international markets and the domestic market, as well.

Este trabalho apresenta um estudo sobre o paradigma de Algoritmos Genéticos aplicados a área de Predições de Séries Temporais. O resultado deste trabalho é apresentado na forma de comparação dos resultados obtidos entre o Modelo Clássico de Predição (UCM), Redes Neurais Artificiais (RNAs) e o modelo de Algoritmos Genéticos desenvolvido neste trabalho. Este estudo foi realizado trabalhando-se basicamente com o Índice Mensal de Produção Industrial do Estado do Rio Grande do Sul fornecido pelo IBGE (Instituto Brasileiro de Geografia e Estatística). Os resultados obtidos mostram que os Algoritmos Genéticos podem atingir níveis satisfatórios de precisão em relação aos valores preditos quando comparados com os valores reais. A validação é feita com predições de um passo à frente e de sete passos à frente. Estas predições são em relação aos sete meses iniciais do ano de 1993. / This work presents a study of Genetic Algorithms paradigm applied to Forecasting Time Series. The results are compared with the obtained with the Classic Model of Prediction (UCM), Artificial Neural Networks (RNAs). This study was accomplished using with the Monthly Index of Industrial Production of the State of Rio Grande do Sul, supplied by the IBGE(Instituto Brasileiro de Geografia e Estatística). The results show that the Genetic Algorithms can accomplish a satisfactory precision when compared with the real values. The validation is made with predictions, one and seven steps ahead. These predictions are equivalent to the seven initial months of 1993.

Este trabalho apresenta um estudo sobre o paradigma de Algoritmos Genéticos aplicados a área de Predições de Séries Temporais. O resultado deste trabalho é apresentado na forma de comparação dos resultados obtidos entre o Modelo Clássico de Predição (UCM), Redes Neurais Artificiais (RNAs) e o modelo de Algoritmos Genéticos desenvolvido neste trabalho. Este estudo foi realizado trabalhando-se basicamente com o Índice Mensal de Produção Industrial do Estado do Rio Grande do Sul fornecido pelo IBGE (Instituto Brasileiro de Geografia e Estatística). Os resultados obtidos mostram que os Algoritmos Genéticos podem atingir níveis satisfatórios de precisão em relação aos valores preditos quando comparados com os valores reais. A validação é feita com predições de um passo à frente e de sete passos à frente. Estas predições são em relação aos sete meses iniciais do ano de 1993. / This work presents a study of Genetic Algorithms paradigm applied to Forecasting Time Series. The results are compared with the obtained with the Classic Model of Prediction (UCM), Artificial Neural Networks (RNAs). This study was accomplished using with the Monthly Index of Industrial Production of the State of Rio Grande do Sul, supplied by the IBGE(Instituto Brasileiro de Geografia e Estatística). The results show that the Genetic Algorithms can accomplish a satisfactory precision when compared with the real values. The validation is made with predictions, one and seven steps ahead. These predictions are equivalent to the seven initial months of 1993.

Submitted by Fabio Sobreira Campos da Costa (fabio.sobreira@ufpe.br) on 2017-02-21T14:53:51Z No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) main_abntex.pdf: 4558296 bytes, checksum: 6f077e7cc7e54787fdfdb3b25b18eabb (MD5) / Made available in DSpace on 2017-02-21T14:53:51Z (GMT). No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) main_abntex.pdf: 4558296 bytes, checksum: 6f077e7cc7e54787fdfdb3b25b18eabb (MD5) Previous issue date: 2016-09-26 / A previsão de séries temporais é uma tarefa importante no campo da aprendizado de máquina, possuindo diversas aplicações em mercado de ações, hidrologia, meteorologia, entre outros. A análise da dependência existente nas observações adjacentes da série é necessária para que seja possível prever valores futuros com alguma precisão. Modelos dinâmicos são utilizados para realizar mapeamentos de uma série temporal, se aproximando do mecanismo gerador da série e sendo capazes de realizar previsões. No entanto, o mecanismo gerador de uma série temporal pode produzir padrões lineares e não-lineares que precisam ser devidamente mapeados. Modelos lineares como o auto-regressivo integrado de média móvel (ARIMA) são capazes de mapear padrões lineares, porém não são indicados quando existem padrões não-lineares na série. Já os modelos não-lineares como as redes neurais artificais (RNA) mapeiam padrões não-lineares, mas podem apresentar desempenho reduzido na presença de padrões lineares em relação aos modelos lineares. Fatores como a definição do número de elementos de entrada da RNA, número de amostras de treinamento podem afetar o desempenho. Abordagens híbridas presentes na literatura realizam o mapeamento dos padrões lineares e não-lineares simultaneamente ou aplicando duas ou mais fases nas previsões. Seguindo a suposição de que os modelos são bem ajustados, a diferença entre o valor previsto e a série real demonstra um comportamento de ruído branco, ou seja, considera-se que a diferença entre os valores (resíduo) é composta por choques aleatórios não correlacionados. Na abordagem de duas ou mais fases, o resíduo gerado pelo modelo aplicado na primeira fase é utilizado pelo segundo modelo. O problema do ajuste pode ser decorrente dos parâmetros mal ajustados e também da série temporal devido à possível necessidade de transformações. Tais abordagens geram previsões mais precisas quando comparadas às técnicas tradicionais. Nesta tese, são explorados sistemas evolucionários para a otimização de parâmetros de técnicas lineares e não-lineares visando o mapeamento dos padrões da série temporal. A abordagem proposta utiliza um preprocessamento automático através de um filtro de suavização exponencial para extrair uma série com distribuição normal. A diferença da série temporal e a série filtrada é mapeada por um sistema composto por um método auto-regressivo (AR) e máquina de vetor de suporte para regressão (SVR). Variações do algoritmo de otimização por enxame de partículas (PSO) e algoritmos genéticos são aplicados na otimização dos hiper-parâmetros do sistema. A previsão final é realizada através da soma das previsões de cada série. Para fins de avaliação do método proposto, experimentos foram realizados com bases de problemas reais utilizando métodos da literatura. Os resultados demonstram que o método obteve previsões precisas na maioria dos casos testados. O filtro de suavização exponencial utilizado supõe que a série possua nível constante (sem tendência). Séries que possuem tendências lineares foram devidamente tratadas, no entanto tendências exponenciais ou polinomiais apresentaram desempenho reduzido. O método proposto possui potencial para melhorias, aplicando métodos que realizem o mapeamento automático de tendências como a suavização exponencial dupla. Nesta tese o método aditivo foi utilizado para combinação de previsões, no entanto em algumas séries o modelo multiplicativo pode ser mais adequado, produzindo previsões mais precisas. / Time series forecasting is an important task in the field of machine learning and has many applications in stock market, hydrology, weather and so on. The analysis of the dependence between adjacent observations in the series is necessary in order to achieve better forecasts. Dynamic models are used to perform mappings in the time series by approximating to thedata generating process and being able to perform predictions. However, the data generating process of a time series may produce both linear and nonlinear patterns that need to be mapped. Linear models such as the autoregressive integrated moving average (ARIMA) are able to map linear patterns, although not indicated when nonlinear patterns are present in the series. Nonlinear models such as the artificial neural networks (ANNs) perform nonlinear mappings but demonstrate reduced performance in the presence of linear patterns in comparison to linear models. Hybrid approaches in the literature perform mappings of linear and nonlinear patterns simultaneously or applying two or more phases.Supposing that the models are adjusted to the data, the difference between the predicted value and the data presents a White noise behavior, thus it is considered that the difference of values (residual) is composed by uncorrelated random shocks. In two-phase approaches the residual produced by the linear model in the first phase is used in the nonlinear model. Also the parameters of the models have an important influence on their performance. Such approaches produce more accurate predictions when compared with traditional methods. In this thesis, we explore evolutionary system in the context of optimization of parameters for both linear and nonlinear methods, taking into consideration the patterns in a time series. In the proposed approach, an exponential smoothing filter is used to decompose a series with normal distribution which is applied to an ARIMA model and the residual series is applied to a system composed by an autoregressive (AR) and a support vector regression methods (SVR). Variations of particle swarm optimization (PSO) algorithm and genetic algorithm (GA) are employed in the optimization of hyper-parameters of the system. Experiments were conducted using data sets from real world problems comparing with methods in the literature. The results indicate that the method achieved accurate predictions in most cases. The exponential smoothing filter assumes that the given series has no trend patterns. Series with linear trend were detrended, however in series with exponential or polynomial trends the proposed method achieved reduced performance. The proposed method has potential to improvements by using methods that perform an automatic mapping of trend patterns (double exponential smoothing). In this work, the additive model is adopted, however in some series a multiplicative model could achieve better forecasts.

[pt] Esta dissertação apresenta uma proposta de arquitetura de redes neurais de intervalos para previsão de séries financeiras. O desempenho desta arquitetura é analisado através de testes de previsão para algumas séries de mercado. Como contribuição adicional é apresentado um algoritmo de trading automático. Este algoritmo é avaliado aplicando-o à séries de mercado, para mensuração de lucros percentuais. Por fim, dados de previsão, obtidos pela rede proposta, são utilizadas para a otimização do trading. / [en] This text presents a new Neural network architeture to be employed in the forecast of financial series. The architecture´s performance is evaluated through benchmarks, using data from financial series. As an additional contribution, an automatic trading algorithm, which is also evaluated through benchmarks, is presented. Finally, forecast data, obtained with the proposed NN architecture, is used to improve the trading algorithm´s performance.

In this master thesis project, four well known time series forecasting models areconstructed and tuned with the purpose of predicting the future consumption of glueon one of AkzoNobels customers production lines. The goal was to examine thepossibility of utilizing their vastly collected data with these models to improve on theinventory management for both AkzoNobel and their customers. The predictedproduct usage rate would aid in the customers' decision making about when neworders of product should be placed, based on when the current storage tanks areforecasted to be emptied. This information could also be useful for AkzoNobelthemselves. The data that is handled in this project is a time series with timestampsfor every glue consumption process on the customers production line since 2017. Asubgoal was to determine what data resolution would be the most effective formodelling, so each model has two versions, one using higher and one using lowerresolution data. The models that are examined are a seasonal naive model,along-short term memory model, a Facebook Prophet model as well as two separateAutoregressive Integrated Moving Average models, specifically one automaticallyandone manually constructed. Beyond these models, a combined model using trueaveraging of the two automatic ARIMA models was examined as well.   Ultimately it was found that, for most models, forecasting ahead with a one day resolution was the most accurate using the models trained on one-day-separated-data, compared to three-hour-separated-data. Further it is presented that the best models are the two naive models, closely followed by the one-day-case automatic ARIMA and Prophet models. These models also performed similarly on simple tests for predicting a date when a tank will be empty. Mostly differing around four days on average from the true date for an empty tank on those tests, with a max forecast range of forty days. It is concluded that it is possible to sufficiently model the data to a point where the best models in this project could be an effective tool for both the AkzoNobel and its customers.