Being in the era of Big data, the applicability and importance of data-driven models like artificial neural network (ANN) in the modern statistics have increased substantially. In this dissertation, our main goal is to contribute to the development and the expansion of these ANN models by incorporating Bayesian learning techniques. We have demonstrated the applicability of these Bayesian ANN models in interdisciplinary research including health and cybersecurity. Breast cancer is one of the leading causes of deaths among females. Early and accurate diagnosis is a critical component which decides the survival of the patients. Including the well known ``Gail Model", numerous efforts are being made to quantify the risk of diagnosing malignant breast cancer. However, these models impose some limitations on their use of risk prediction. In this dissertation, we have developed a diagnosis model using ANN to identify the potential breast cancer patients with their demographic factors and the previous mammogram results. While developing the model, we applied the Bayesian regularization techniques (evidence procedure), along with the automatic relevance determination (ARD) prior, to minimize the network over-fitting. The optimal Bayesian network has 81\% overall accuracy in correctly classifying the actual status of breast cancer patients, 59\% sensitivity in accurately detecting the malignancy and 83\% specificity in correctly detecting non-malignancy. The area under the receiver operating characteristic curve (0.7940) shows that this is a moderate classification model. We then present a new Bayesian ANN model for developing a nonlinear Poisson regression model which can be used for count data modeling. Here, we have summarized all the important steps involved in developing the ANN model, including the forward-propagation, backward-propagation and the error gradient calculations of the newly developed network. As a part of this, we have introduced a new activation function into the output layer of the ANN and error minimizing criterion, using count data. Moreover, we have expanded our model to incorporate the Bayesian learning techniques. The performance our model is tested using simulation data. In addition to that, a piecewise constant hazard model is developed by extending the above nonlinear Poisson regression model under the Bayesian setting. This model can be utilized over the other conventional methods for accurate survival time prediction. With this, we were able to significantly improve the prediction accuracies. We captured the uncertainties of our predictions by incorporating the error bars which could not achieve with a linear Poisson model due to the overdispersion in the data. We also have proposed a new hybrid learning technique, and we evaluated the performance of those techniques with a varying number of hidden nodes and data size. Finally, we demonstrate the suitability of Bayesian ANN models for time series forecasting by using an online training algorithm. We have developed a vulnerability forecast model for the Linux operating system by using this approach.

The Time Dependent Kernel Density Estimation (TDKDE) developed by Harvey & Oryshchenko (2012) is a kernel density estimation adjusted by the Exponentially Weighted Moving Average (EWMA) weighting scheme. The Maximum Likelihood Estimation (MLE) procedure for estimating the parameters proposed by Harvey & Oryshchenko (2012) is easy to apply but has two inherent problems. In this study, we evaluate the performances of the probability density estimation in terms of the uniformity of Probability Integral Transforms (PITs) on various kernel functions combined with different preset numbers. Furthermore, we develop a new estimation algorithm which can be conducted using Artificial Neural Networks to eliminate the inherent problems with the MLE method and to improve the estimation performance as well. Based on the new estimation algorithm, we develop the TDKDE-based Random Forests time series classification algorithm which is significantly superior to the commonly used statistical feature-based Random Forests method as well as the Ker- nel Density Estimation (KDE)-based Random Forests approach. Furthermore, the proposed TDKDE-based Self-organizing Map (SOM) clustering algorithm is demonstrated to be superior to the widely used Discrete-Wavelet- Transform (DWT)-based SOM method in terms of the Adjusted Rand Index (ARI).

>Magister Scientiae - MSc / This study has been conducted to accurately determine fluid saturation within Fula sub-basin reservoirs which is located at the Southern part of the Republic of Sudan. The area is regarded as Shaly Sand Reservoirs. Four deferent shaly sand lithofacies (A, B, C, D) have been identified. Using method based on the Artificial Neural Networks (ANN), the core surrounding facies, within Fula reservoirs were identified. An average shale volume of 0.126 within the studied reservoirs was determined using gamma ray and resistivity logs. While average porosity of 26.7% within the reservoirs was determined using density log and the average core grain density. An average water resistivity of 0.8 Ohm-m was estimated using Pickett plot method. While formation temperature was estimated using the gradient that constrained between surface and bottom hole temperature. Water saturation was determined using Archie model and four shaly sand empirical models, the calculation was constrained within each facies zone to specify a model for each facies, and another approach was used to obtain the water saturation based on Artificial Neural Networks. The net pay was identified for each reservoir by applying cut-offs on permeability 5 mD, porosity 16%, shale volume 0.33, and water saturation 0.65. The gross thickness of the reservoirs ranges from 7.62m to 19.85m and net pay intervals from 4.877m to 19.202m. The study succeeded in establishing water saturation model for the Fula sub-basin based on neural networking which was very consistent with the core data, and hence has been used for net pay determination.

The need for a more authoritative approach to investment decision-making and cost control has been a requirement of office spending for many years now. The commercial offices find itself in an increasingly demanding position to allocate its budgets as wisely and prudently as possible. The significant percentage of total spending on buildings demands a more accurate and adaptable method of achieving quality of service within the constraints on the budgets. By adoption of life cycle costing techniques with risk management, practitioners have the ability to make accurate forecasts of likely future running costs. This thesis presents a novel framework (Artificial Neural Networks and probabilistic simulations) for modelling of operating and maintenance historical costs as well as economic performance measures of LCC. The methodology consisted of eight steps and presented a novel approach to modelling the LCC of operating and maintenance costs of two sustainable commercial office buildings. Finally, a set of performance measurement indicators were utilised to draw inference from these results. Therefore, the contribution that this research aimed to achieve was to develop a dynamic LCC framework for sustainable commercial office buildings, and by means of two existing buildings, demonstrate how assumption modelling can be utilised within a probabilistic environment. In this research, the key themes of risk assessment, probabilistic assumption modelling and stochastic assessment of LCC has been addressed. Significant improvements in existing LCC models have been achieved in this research in an attempt to make the LCC model more accurate and meaningful to estate managers and high-level capital investment decision makers A new approach to modelling historical costs and forecasting these costs in sustainable commercial office buildings is presented based upon a combination of ANN methods and stochastic modelling of the annual forecasted data. These models provide a far more accurate representation of long-term building costs as the inherent risk associated with the forecasts is easily quantifiable and the forecasts are based on a sounder approach to forecasting than what was previously used in the commercial sector. A novel framework for modelling the facilities management costs in two sustainable commercial office buildings is also presented. This is not only useful for modelling the LCC of existing commercial office buildings as presented here, but has wider implications for modelling LCC in competing option modelling in commercial office buildings. The processes of assumption modelling presented in this work can be modified easily to represent other types of commercial office buildings. Discussions with policy makers in the real estate industry revealed that concerns were held over how these building costs can be modelled given that available historical data represents wide spending and are not cost specific to commercial office buildings. Similarly, a pilot and main survey questionnaire was aimed at ascertaining current level of LCC application in sustainable construction; ranking drivers and barriers of sustainable commercial office buildings and determining the applications and limitations of LCC. The survey result showed that respondents strongly agreed that key performance indicators and economic performance measures need to be incorporated into LCC and that it is important to consider the initial, operating and maintenance costs of building when conducting LCC analysis, respondents disagreed that the current LCC techniques are suitable for calculating the whole costs of buildings but agreed that there is a low accuracy of historical cost data.

This thesis focuses on creation and comparison of ice hockey matches prediction models with the view on ice hockey world championship matches. The first part is dedicated to collecting theoretical knowledge needed for solving this problem and the second to applying this set of knowledge. The model creation approach is intertwined with the CRISP-DM data mining methodology, which also defines several chapters of this work. As input data for the models I used performance statistics of individual ice hockey players -- this brought me to implementing a script capable of automatic downloading and aggregating of player data from the Internet. Downloaded data were arranged so as they would represent ice hockey matches that were played during the championships (team A consisting of players X against team B consisting of players Y) with result of the match added to the data row. Data were also analyzed to detect any quality issue prior to the model creation and transformed into an integrated view. Result assessment consists of two parts, in the first the technical evaluation of models using data from the testing data set takes place. The first part also points out practical usefulness of the models. The next part is about comparing result data with the betting odds -- the business relevance of the model. This part uses open source data about betting odds listed on the corresponding matches. Finally, the outcome model is used for predicting matches of the group phase of the world championship taking place in Prague, 2015.

Finding significant relation between stock markets (including omnipresent volatility) and real economy of the US, Germany, Great Britain and Japan is the main aim of this thessis. If not found it is also the final conclusion. By means of time series analysis using artificial neural networks from the beginning of 2000 till the November of 2014 was proved that the strong single -- way relation between prime stock indices and GDP of chosen economies does exist. Highest quality of prediction was proved on the American and British economy. S&P 500, FTSE and VIX indicator made a precise prediction of future economic progress in the US and Great Britain for six to nine months ahead with 71% to 86% accuracy. The artificial neural networks proved an extraordinary ability to predict chosen financial time series regardless the actual position in a business cycle.

O cérebro humano é formado por um conjunto de neurônios de diferentes tipos, cada um com sua especialidade. A combinação destes diferentes tipos de neurônios é um dos aspectos responsáveis pelo desempenho apresentado pelo cérebro na realização de várias tarefas. Redes Neurais Artificiais são técnicas computacionais que apresentam um modelo matemático inspirado no sistema nervoso e que adquirem conhecimento através da experiência. Uma alternativa para melhorar o desempenho das Redes Neurais Artificiais é a utilização de técnicas de Combinação de Classificadores. Estas técnicas de combinação exploram as diferenças e as semelhanças das redes para a obtenção de resultados melhores. Dentre as principais aplicações de Redes Neurais Artificiais está o Reconhecimento de Padrões. Neste trabalho, foram utilizadas técnicas de Combinação de Classificadores para a combinação de Redes Neurais Artificiais em problemas de Reconhecimento de Padrões. / The human brain is formed by neurons of different types, each one with its own speciality. The combination of theses different types of neurons is one of the main features responsible for the brain performance in severa! tasks. Artificial Neural Networks are computation technics whose mathematical model is based on the nervous system and learns new knowledge by experience. An alternative to improve the performance of Artificial Neural Networks is the employment of Classifiers Combination techniques. These techniques of combination explore the difference and the similarity of the networks to achieve better performance. The main application of Artificial Neural Networks is Pattern Recognition. In this work, Classifiers Combination techniques were utilized to combine Artificial Neural Networks to solve Pattern Recognition problems.

Este trabalho descreve experimentos realizados com Redes Neurais Artificiais e algoritmos de aprendizado simbólico. Também são investigados dois algoritmos de extração de conhecimento de Redes Neurais Artificiais. Esses experimentos são realizados com três bases de dados com o objetivo de comparar os desempenhos obtidos. As bases de dados utilizadas neste trabalho são: dados de falência de bancos brasileiros, dados do jogo da velha e dados de análise de crédito. São aplicadas sobre os dados três técnicas para melhoria de seus desempenhos. Essas técnicas são: partição pela menor classe, acréscimo de ruído nos exemplos da menor classe e seleção de atributos mais relevantes. Além da análise do desempenho obtido, também é feita uma análise da dificuldade de compreensão do conhecimento extraído por cada método em cada uma das bases de dados. / This work describes experiments carried out witch Artificial Neural Networks and symbolic learning algorithms. Two algorithms for knowledge extraction from Artificial Neural Networks are also investigates. This experiments are performed whit three data set with the objective of compare the performance obtained. The data set used in this work are: Brazilians banks bankruptcy data set, tic-tac-toe data set and credit analysis data set. Three techniques for data set performance improvements are investigates. These techniques are: partition for the smallest class, noise increment in the examples of the smallest class and selection of more important attributes. Besides the analysis of the performance obtained, an analysis of the understanding difficulty of the knowledge extracted by each method in each data bases is made.

<p>Predmet ove doktorske disertacije je analiza dinamičkog ponašanja<br />kotrljajnih ležaja primenom veštačkih neuronskih mreža. Na bazi<br />rezultata eksperimentalnog ispitivanja obučene su veštačke<br />neuronske mreže koje su sposobne da predvide amplitude brzine<br />vibracija ležaja. Vibracije koje ležaj generiše zavise od niza<br />uticajnih parametara koji se mogu podeliti na konstrukcione,<br />tehnološke i eksploatacione. Modeli dobijeni primenom veštačkih<br />neuronskih mreža određuju zavisnosti između uticajnih parametara i<br />amplituda brzine vibracija koje ležaj generiše. Validacija<br />neuronskih modela izvršena je na osnovu eksperimentalnih rezultata.<br />Analiziran je uticaj svakog parametra ležaja na amplitude brzine<br />vibracija u karakterističnim područjima frekvencija. U radu su<br />prikazani i rezultati međusobnog uticaja više parametara. Modelima<br />su dobijene preporučene vrednosti uticajnih parametara ležaja. Pri<br />analizi tehnoloških parametara uvedeni su: parametar ekvivalentne<br />površinske hrapavosti, parametar ekvivalentne valovitosti i<br />parametar ekvivalentnog odstupanja od kružnosti staza kotrljanja.<br />Novouvedeni parametri omogućavaju bolje razumevanje uticaja na<br />dinamičko ponašanje. U radu je pokazano da su neuronski modeli<br />sposobni da na osnovu parametara ležaja predvide klasu kvaliteta<br />ležaja.</p> / <p>The subject of this doctoral dissertation is the analysis of the dynamic<br />behavior of ball bearings using artificial neural networks. Based on the<br />results of the experimental test, artificial neural networks were trained to be<br />able to predict the amplitudes of the bearing vibration velocity. The vibrations<br />generated by the bearing depend on a number of influential parameters that<br />can be divided into construction, technological and exploitation. Models<br />obtained by applying artificial neural networks determined the dependences<br />between the influencing parameters and the amplitudes of the vibration<br />velocity generated by the bearing. Validation of neural models was<br />performed based on experimental results. The influence of each parameter<br />on the vibration velocity amplitudes in the characteristic frequency ranges<br />was analyzed. The paper also presents the results of the mutual influence of<br />several parameters. The models obtained the recommended values of the<br />influential bearing parameters. In the analysis of technological parameters,<br />the following parameters were introduced: the parameter of equivalent<br />surface roughness, the parameter of equivalent waviness and the parameter<br />of equivalent roundness error of raceways. The newly introduced parameters<br />provide a better understanding of the impact on dynamic behavior. The paper<br />shows that neural models are able to predict the bearing quality class based<br />on bearing parameters.</p>

<p>Problem velike količine emisije CO2 u atmosferi je međunarodno<br />prepoznat a Evropska unija je dokumentom &bdquo;Energetska mapa puta 2050<br />Evropske unije&ldquo; najviše doprinela u prepoznavanju i realizaciji mera<br />za njegovo sprovođenje. Jedan od ključnih segmenata dokumenta<br />predstavlja energetska politika. U ovoj tezi su prepoznati ključni<br />indikatori vezani za energetsku politiku, a zatim je formiran model<br />veštačkih neuronskih mreža koji je u stanju da predvidi emisiju CO2<br />do 2050. godine. Model je u mogućnosti da nauči funkcionisanje celog<br />sistema i omogućava simulaciju raznih scenarija energetske politike<br />kako bi se ispunio cilj da se što efikasnije i brže dođe do željenog<br />smanjenja emisija CO2.</p> / <p>The problem of high CO2 emission has been recognised internationally.<br />European Union has contributed the most to recognition and realization of<br />measures and actions, needed to solve this problem, by developing<br />document Energy Roadmap 2050. One of key segments of this document is<br />energy policy. In this thesis, key indicators for energy policy are found, after<br />which the artificial neural network model which is capable of CO2 emission<br />prediction by the year 2050 is formed. Model is capable of learning the whole<br />complex energy system and enables simularion of different energy policy<br />scenarios in order to reach the EU goal and decrease CO2 emission by the<br />year 2050 in most efficient and easiest way.</p>