A Linear mixed-effects model (LME) is one of the possible tools for longitudinal or group--dependent data. This thesis deals with evaluating of prediction error in LME. Firstly, it is derived the mean square error of prediction (MSEP) by direct calculation. Then the covariance penalty method and crossvalidation is presented for evaluation of MSEP in LME. Further, it is shown how Akaike information criterion (AIC) can be used in mixed-effects models. Because of the model's properties two types of AIC are distinguished - marginal and conditional one. Subsequently, the procedures of AIC's calculation and its basic asymptotic properties are described. Finally, the thesis contains simulation study of behaviour of marginal and conditional AIC with the goal to choose the right variance structure of random effects. It turns out that the marginal criterion tends to select models with smaller number of random effects than conditional criterion.

Mesoscale design involving control of material distribution pattern can create a statistically heterogeneous material system, which has shown increased adaptability to complex mechanical environments involving highly non-uniform stress fields. Advances in multi-material additive manufacturing can aid in this mesoscale design, providing voxel level control of material property. This vast freedom in design space also unlocks possibilities within optimization of the material distribution pattern. The optimization problem can be divided into a forward problem focusing on accurate predication and an inverse problem focusing on efficient search of the optimal design. In the forward problem, the physical behavior of the material can be modeled based on fundamental mechanics laws and simulated through finite element analysis (FEA). A major limitation in modeling is the unknown parameters in constitutive equations that describe the constituent materials; determining these parameters via conventional single material testing has been proven to be insufficient, which necessitates novel and effective approaches of calibration. A calibration framework based in Bayesian inference, which integrates data from simulations and physical experiments, has been applied to a study involving a mesostructured material fabricated by fused deposition modeling. Calibration results provide insights on what values these parameters converge to as well as which material parameters the model output has the largest dependence on while accounting for sources of uncertainty introduced during the modeling process. Additionally, this statistical formulation is able to provide quick predictions of the physical system by implementing a surrogate and discrepancy model. The surrogate model is meant to be a statistical representation of the simulation results, circumventing issues arising from computational load, while the discrepancy is aimed to account for the difference between the simulation output and physical experiments. In this thesis, this Bayesian calibration framework is applied to a material bending problem, where in-situ mechanical characterization data and FEA simulations based on constitutive modeling are combined to produce updated values of the unknown material parameters with uncertainty. / Master of Science / A material system obtained by applying a pattern of multiple materials has proven its adaptability to complex practical conditions. The layer by layer manufacturing process of additive manufacturing can allow for this type of design because of its control over where material can be deposited. This possibility then raises the question of how a multi-material system can be optimized in its design for a given application. In this research, we focus mainly on the problem of accurately predicting the response of the material when subjected to stimuli. Conventionally, simulations aided by finite element analysis (FEA) were relied upon for prediction, however it also presents many issues such as long run times and uncertainty in context-specific inputs of the simulation. We instead have adopted a framework using advanced statistical methodology able to combine both experimental and simulation data to significantly reduce run times as well as quantify the various uncertainties associated with running simulations.

This thesis deals with a real application of econometric methods to the analysis of electric energy consumption in a significant Czech brewery. The main objective is to construct a model predicting the electric energy consumption in the production process in the next week based on various data measured in the last 2 years. Results will be used in the costs management of the company.

This thesis deals with comparing two approaches to modelling and predicting time series: a traditional one (the ARIMAX model) and a modern one (gradiently boosted decision trees within the framework of the XGBoost library). In the first part of the thesis we introduce the theoretical framework of supervised learning, the ARIMAX model and gradient boosting in the context of decision trees. In the second part we fit the ARIMAX and XGBoost models which both predict a specific time series, the daily volume of the S&P 500 index, which is a crucial task in many branches. After that we compare the results of the two approaches, we describe the advantages of the XGBoost model, which presumably lead to its better results in this specific simulation study and we show the importance of hyperparameter optimization. Afterwards, we compare the practicality of the methods, especially in regards to their computational demands. In the last part of the thesis, a hybrid model theory is derived and algorithms to get the optimal hybrid model are proposed. These algorithms are then used for the mentioned prediction problem. The optimal hybrid model combines ARIMAX and XGBoost models and performs better than each of the individual models on its own. 1

The thesis deals with the design of a permanent magnet synchronous machine controller that isimplemented on an embedded platform to replace the off-the-shelf controller currently being used in theelectric race car of the KTH Formula Student team. Software implementation of the control algorithmwas tested in laboratory environment on the hardware prototype of a 2-level three-phase voltage sourceinverter.Field oriented control and finite control set model predictive control algorithms were implemented insimulation environment. The latter performed better in terms of reducing switching activity and torqueripple, but needs vastly more computational resources due to its nature of being an online optimizationproblem. Trade-off curve of phase current harmonic distortion and switching activity showed that themodel prediction control algorithm performs better in the low frequency range (1-20 kHz). Obtainedsimulation results were used for power electronics component selection.Field oriented control was implemented on a TMS320F28335 DSP. SPI communication was employedto configure gate driver circuits and perform error handling. The DSP program follows interrupt basedorganization and the main control loop runs on the variable frequency of the pulse width modulation.Low voltage test results on three-phase inductive-resistive load showed that the controller outputssinusoidal current. Efficiency measurement, high voltage and motor testing were hindered by interferencefrom the Silicon-Carbide MOSFETs that prohibited correct operation of hardware. / Den här uppsatsen handlar om designen och implementeringen av en motorstyrning för en permanen- magnetiserad synkronmotor, med syfte att ersätta standardmotorstyrningsenheten i KTH Formula Students tävlingsbil. Implementationen av styralgoritmen testades experimentellt tillsammans med en prototyptillverkad frekvensomriktare i labbmiljö. Regleralgoritmer för field oriented control och finite control set model predictive control implementerades och testades i simuleringsmiljö. Den senare algoritmen visade sig prestera bättre i form av lägre vridmomentsoscillationer trots lägre switch-frekvens men den kräver samtidigt mer beräkningskraft. Övertonsinnehållet (THD) i fasströmmarna som funktion av switchfrekvensen undersöktes för de båda regleralgoritmerna, algoritmen för model predictive control gav lägre THD vid lägre frekvenser (1-20 kHz). Simuleringsresultaten användes för att motivera valet av komponenter till frekvensomriktaren. Regleralgoritmen för field oriented control implementerades och testades experimentellt med hjälp av ett utvecklingskort (TMS320F28335) från Texas Instruments. SPI-kommunikation användes för att konfigurera drivkretsana samt för att utläsa felkoder. Experimentalla tester som utfördes på låg spänningsnivå visade att strömmen till lasten var sinusformad. Mätning av verkningsgrad och provning tillsammans med motorn på en högre spänningsnivå gick inte att geno av att de snabba switchförloppen i kiselkarbidtransistorerna störde ut motorstyrningen.

<p>Ekstremni vidovi ispoljavanja agresije u vidu krivičnih dela nasilja su univerzalni<br />fenomeni, prepoznati u svim dru&scaron;tvima i kulturama. Iako se radi o relativno nefrekventnim<br />događajima, trend nasilničkog pona&scaron;anja raste i predstavlja problem od &scaron;ireg dru&scaron;tvenog značaja.<br />Ishod vi&scaron;edecenijskog izučavanja agresivnosti ogleda se u detektovanju velikog broja prediktora<br />agresivnog pona&scaron;anja, pri čemu su se faktori agresivnosti najče&scaron;će izučavali izolovano. Kao<br />referentni okvir za ovo istraživanje poslužio nam je Op&scaron;ti model agresivnosti koji podrazumeva<br />međusobnu interakciju različitih faktora u generisanju agresivnog pona&scaron;anja. Osnovni cilj ovog<br />istraživanja se odnosi na rasvetljavanje glavnih i interaktivnih efekta prediktora, a &scaron;to doprinosi<br />boljem razumevanju uslova pod kojima se povećava ili smanjuje verovatnoća realizacije<br />agresivnog pona&scaron;anja, u kontekstu krivičnih dela.<br />Istraživanjem je obuhvaćeno 200 ispitanika, mu&scaron;kog pola, podeljenih u dve grupe.<br />Kriterijsku grupu činilo je 100 ispitanika koji su bili na izdržavanju kazne u Kazneno popravnom<br />zatvoru u Sremskoj Mitrovici, zbog krivičnog dela nasilja. Kontrolnu grupu činilo je 100<br />ispitnanika koji u svojoj istoriji nisu imali izvr&scaron;eno ni jedno krivično delo. Ispitanici su<br />ujednačeni u odnosu na psihijatrijsku dijagnozu.<br />Organizovan je korelacioni nacrt, a rezultati su obrađeni kroz transferzalnu perspektivu.<br />Ispitivanje interaktivnog uticaja prediktorskih varijabli sprovedeno je putem ispitivanja<br />moderacije.<br />Rezultati istraživanja pokazuju da je na osnovu personolo&scaron;ko-dispozicionih, kognitivnoemocionalnih<br />i socio-emografskih prediktora moguće razlikovati grupu počinilaca krivičnog dela<br />nasilja u odnosu na ispitanike koji nikada nisu počinili krivično delo. Konkretnije, grupu<br />počinilaca krivičnih dela nasilja karakteri&scaron;e vi&scaron;i stepen izraženosti sve tri Eysenck-ove<br />dimenzije, vi&scaron;i stepen sklonosti ka agresivnom pona&scaron;anju, vi&scaron;i stepen izraženosti psihopatske<br />devijacije, dok se po pitanju stepena samopo&scaron;tovanja ne razlikuju u odnosu na kontrolnu grupu.<br />Počinioci krivičnog dela nasilja imaju specifičan kognitivni stil koji je obeležen većim<br />prisustvom agresivnih fantazija, neefikasnom kontrolom agresivnih i neprijatnih misli, većim<br />prisustvom iracionalnih uverenja, uz če&scaron;će kori&scaron;ćenje supresije, kao neadekvatne strategije<br />vi<br />emocionalne regulacije. Takođe, počinioci krivičnog dela potiču iz porodica sa većim stepenom<br />alkoholizma (isključivo oca), u većoj meri su bili izloženi nasilju u formativnom periodu, imaju<br />niži stepen obrazovanja i ređe imaju stalno zaposlenje. Međutim, kada se ovi brojni faktori<br />agresivnosti posmatraju kroz prizmu multivarijatnog modela predikcije, samo mali broj ostvaruje<br />glavni prediktivni doprinos. Izdvojili su se sledeći prediktori: sklonost ka antisocijalnom<br />pona&scaron;anju, samopo&scaron;tovanje, netolerancija životnih događaja, supresija, reprocenjivanje i<br />obrazovni status. Ispitujući interaktivan efekat prediktorskih varijabili i psihopatije, kao<br />moderator varijable, rezultati pokazuju da različit stepen izraženosti subdimenzija psihopatije<br />predstavlja uslov pod kojim personolo&scaron;ko-dispozicione varijable ostvaruju značajan doprinos u<br />prdikciji agresivnog pona&scaron;anja.<br />Uzimajući u obzir da su se kognitivno-emocionalni faktori koji su podložni promeni<br />izdvojili kao značajni prediktori, praktičan cilj istraživanja ogleda se u primeni nalaza<br />istraživanja na proces rehabilitacije počinilaca agresivnih krivičnih dela</p> / <p>Extreme forms of aggression manifestations, in terms of violent crimes, are universal<br />phenomena recognized in all societies and cultures. Although these are relatively small<br />frequency events, the trend of violent behaviour is growing and represents a problem of wider<br />social significance. The result of multiple decades researches of human aggression is the<br />detection of a large number of aggressive behaviour predictors, where the aggression factors<br />were most commonly studied as isolated ones. As a reference framework for this research, the<br />General Aggression Model was used, as it comprehends different factors mutual interaction in<br />generation of aggressive behaviour. The basic aim of this research is to put some more light to<br />the main and interactive predictor effects, which contributes to better understanding of the<br />conditions under which the probability of realization of the aggressive behaviour is rising or<br />lowering, in terms of criminal acts.<br />The research was performed on 200 male subjects divided into two groups. The criteria<br />group was formed out of 100 subjects who are imprisoned in Sremska Mitrovica Penitentiary for<br />violent crimes. The control group was formed out of 100 subjects who do not have a criminal<br />history at all. The subjects are uniform with relation to psychiatric diagnosis.<br />Correlation design was organized and the results were processed through transversal<br />perspective. Examination of the interactive influence of the predictor variables was performed<br />through moderation.<br />Research results show that it is possible to distinguish the group of violence offenders<br />from the group of subjects with no criminal history at all, on the basis of personologicaldispositional,<br />cognitive-emotional and socio-demographical predictors. Specifically, the group of<br />violent offenders is characterized by the higher level of expression of all three Eysenck<br />dimensions, higher level of inclination to aggressive behaviour, higher level of expression of<br />psychopathic deviation, while the level of self-esteem is no different to the control group.<br />Violent offenders have a specific cognitive style which is marked by higher presence of<br />aggressive fantasies, non-efficient control of aggressive and unwanted thoughts, higher presence<br />of irrational beliefs, with more frequent use of suppression as inadequate strategy for emotional<br />viii<br />regulation. Also, violent offenders come from families with higher level of alcoholism (father<br />only), they have been exposed to violence to a bigger extent in their formation period, they have<br />a lower level of education and less frequently have a permanent employment. However, when<br />these numerous factors of aggression are observed through the prism of multivariate model of<br />prediction, only a small number of factors realize the main predictive effect. The following<br />predictors are noted as significant: inclination to antisocial behaviour, self-esteem, low<br />frustration tolerance beliefs, suppression, reappraisal and educational status. Examination of<br />interactive effect of predictor variables and psychopathy, as moderator variable, gives results<br />which show that different degree of expression of psychopathy sub-dimensions represents the<br />condition under which the personological-dispositional variables give significant contribution to<br />aggressive behaviour prediction.<br />Taking into account that the cognitive-emotional factors which are subject of change are<br />shown to be significant predictors, the practical aim of this research is to apply the research<br />results in violent offenders&rsquo; rehabilitation process.</p>

Production of and demand for Canadian canola meal have been increased yearly. In order to improve the competitiveness of canola meal domestically and internationally, as well as to develop potential markets for canola meal, it is necessary to develop canola meal-based products that have high feed values and can be easily transported. The objectives of this research were: 1) to investigate the effects of temperature and time of conditioning during pelleting process on the nutritive values of canola meal in terms of chemical profiles, protein and carbohydrate subfractions, and energy values, using the AOAC procedures, CNCPS v6.1 and NRC (2001), respectively; 2) to detect the effects of temperature and time of conditioning during the pelleting process on rumen degradation and intestinal digestion characteristics and predicted protein supply of canola meal, using the in situ procedure, the three-step in vitro procedure, and the NRC 2001 model; and 3) to determine pelleting-induced changes in spectral characteristics of molecular structures of canola meal using Attenuated Total Reflectance Fourier Transform Infrared Spectroscopy (ATR-FTIR) with univariate and multivariate analysis, and reveal the relationship between molecular structures of protein and carbohydrate and nutrient values, rumen degradation and intestinal digestion characteristics, and predicted protein supply of canola meal. Three different conditioning temperatures (70, 80 and 90ºC) and two different conditioning time (50 and 75 sec) were applied in this research. Two different batches of canola meal from a commercial feed company were selected. A randomized complete block design (RCBD) with 3 × 2 factorial arrangement was employed in this research. Molecular spectral functional groups related to protein, cellulosic compounds, and carbohydrates were used in the spectral study. This research indicated: 1) soluble crude protein (SCP) was decreased and neutral detergent insoluble CP (NDICP) was increased with increasing temperature; 2) the lowest protein rumen degradation of pellets was observed at conditioning temperature of 90 ºC and protein rumen degradation was increased by pelleting; 3) the amount of protein digested in the small intestine tended to increase with increasing conditioning temperature; 4) pelleting under different temperatures and time in the current study shifted the protein digestion site to the rumen, rather than to the small intestine; 5) with respect to predicted protein supply, based on the NRC 2001 model, increasing conditioning temperature tended to increase the metabolizable protein supply of canola meal pellets to dairy cattle; 6) changes in the molecular structure of canola meal induced by pelleting can be detected by ATR-FTIR; 7) not only protein molecular structure characteristics but also carbohydrate molecular structure characteristics play important roles in determining nutrient values, rumen degradation and intestinal digestion characteristics, and the predicted protein supply of canola meal.

Consumer function is a standard instrument of quantitative economic analysis to examine the relationship between consumer expenditure and income or other influencing factors such as liquid assets, interest rates or various demographic and social factors. In this thesis are presented the most frequently used methods in econometric analysis of consumption function. Attention is paid to the hypothesis of absolute income, relative income, life cycle, permanent income, rational expectations and consumption function based on the error correction model. Furthermore, the suitability of individual models for the current economic situation in the Czech Republic is assessed. Subsequently an empirical model of consumption function for the Czech Republic is designed and tested. Furthermore, the estimates of each consumption function model for the period before and after economic crisis of 2008-2009 are performed and compared. Finally, a short-term prediction of the consumption of Czech households is made.

Determining the prognosis or risk of an individual experiencing a specific health outcome within a certain time period is essential to improve health. An important aspect of prognostic research is the development of risk assessment models (RAMs). In support of the movement towards personalized medicine, health care professionals have employed RAMs to stratify an individual patient’s absolute risk of developing a health condition and select the optimal management strategy for that patient. The development of RAMs is generally conducted using data driven methods or through expert consensus. However, these methods present limitations. Accordingly, we recognized the need to select factors for RAM development or update that are evidence-based and clinically relevant using a structured and transparent approach. In this sandwich thesis, I highlight the methods used to select prognostic factors for VTE and bleeding RAMs for hospitalized medical patients. However, the same methods can be applied to any clinical outcome of interest. This work presents a conceptualized and tested novel mixed methods approach to select prognostic factors for VTE and bleeding in hospitalized medical patients that are evidence-based, clinically meaningful and relevant. Our findings may inform the development of new RAMs, the update of widely used RAMs, and external validation and prospective impact assessment studies. Also, these findings may assist decision makers in evaluating the risk of an individual having an outcome to optimize patient care. / Thesis / Doctor of Philosophy (PhD) / Measuring the probability of an individual experiencing a specific health outcome in a certain period of time based on that individual’s risk factors is important to improve health. Prediction tools are often used to calculate the probability of an outcome. Health care practitioners use prediction tools to assess an individual’s risk of a certain health outcome and in turn provide individualized management. Prediction tools include a number of agreed upon risk factors that should be assessed in order to best estimate the risk of an outcome. These risk factors are usually selected through exploring sets of data or by consulting a group of experts in the field. However, these methods have limitations. Therefore, we recognized that it is important, when developing prediction tools, to select risk factors that are evidence-based and clinically relevant by adopting a systematic, comprehensive, structured and transparent approach. These sets of risk factors can then aid health researchers when developing new prediction tools or updating existing ones and help clinicians predicting risk. In this thesis, I highlight the methods used to select factors for prediction tools that evaluate the risk of having a venous clot or a bleeding event in patients that are hospitalized for a medical condition. However, the same methods can be applied to any clinical condition and outcome of interest. This work presents a new approach that we conceptualized and tested to select risk factors for venous clots and bleeding events in hospitalized medical patients that are evidence-based, clinically meaningful and relevant. Our findings may inform the development of new prediction tools, the update of widely used tools, and the design of studies to validate these tools. Also, these findings may assist decision makers in evaluating the risk of an individual having an outcome to optimize patient care.

<p>U okviru disertacije, kori&scaron;ćenjem ve&scaron;tačkih neuronskih mreža razvijeni su modeli za predviđanje količina ambalažnog i biorazgradivog komunalnog otpada u Republici Srbiji do kraja 2030. godine. Razvoj modela baziran je na zavisnosti između ukupne potro&scaron;nje domaćinstva i generisane količine dva posmatrana toka otpada. Pored toga, na bazi zavisnosti sa bruto domaćim proizvodom (BDP), definisan je i model za projekciju zastupljenosti osnovnih opcija tretmana komunalnog otpada u Republici Srbiji za isti period. Na osnovu dobijenih rezultata, stvorene su polazne osnove za procenu potencijala za reciklažu ambalažnog otpada, kao i za procenu u kojoj meri se može očekivati da određene količine biorazgradivog otpada u narednom periodu ne budu odložene na deponije, &scaron;to je u skladu sa savremenim principima upravljanja otpadom i postojećim zahtevima EU u ovoj oblasti.</p> / <p>By using artificial neural networks, models for prediction of the quantity of<br />packaging and biodegradable municipal waste in the Republic of Serbia by<br />the end of 2030, were developed. Models were based on dependence<br />between total household consumption and generated quantities of two<br />observed waste streams. In addition, based on dependence with the Gross<br />Domestic Product (GDP), a model for the projection of share of different<br />municipal solid waste treatment options in the Republic of Serbia for the same<br />period, was created. Obtained results represent a starting point for assessing<br />the potential for recycling of packaging waste, and determination of<br />biodegradable municipal waste quantities which expected that in the future<br />period will not be disposed at landfills, in accordance with modern principles<br />of waste management and existing EU requirements in this area.</p>