This thesis compares and evaluates the accuracy of two RSSI-based tri-lateration methods in an indoor setting, implementing either Wi-Fi andBluetooth Low Energy (BLE) while using commercially available hardware.The purpose of evaluation is part of the long-term vision of improving thesafety of workers in adverse environments such as factories, by providing awearable Indoor Positioning System where other systems like GPS are notsuitable due to signal obstruction. Within a confined space replicating in-tended real-world conditions in terms of signal attenuation and adversity,30 consecutive measurements of signal strength readings (RSSI) to threereference nodes were collected at 10 randomized sample positions, andwas repeated across 5 tests. The accuracy of trilateration was evaluatedusing an averaged Root Mean Square Error (RMSE) over the five tests. Itwas observed that RSSI using Wi-Fi achieved better accuracy of predictingthe actual position within the testing environment than signal-strength us-ing BLE, with Wi-Fi and BLE achieving an accuracy of 0.88 and 1.85 metersrespectively. However, because of the power efficiency of BLE it is a viablecandidate for a future low-cost and device-based Indoor Localization Sys-tem to potentially be used and worn by workers. The results while alignedwith similar existing literature, infer what a low-cost indoor positioningsystem might achieve. Future research with the goal of developing suchsolutions could benefit from implementing both Wi-Fi and BLE as the basisof signal strength trilateration.

Maskininlärning (ML) är idag ett mycket aktuellt, populärt och aktivt forskat område. Därav finns det idag en stor uppsjö av olika avancerade och moderna ML-algoritmer. Svårigheten är att bland dessa identifiera den mest optimala att applicera på ens tillämpningsområde. Algoritmer som bygger på Gradientförstärkning (eng. Gradient Boosting (GB)) har visat sig ha ett väldigt brett spektrum av appliceringsområden, flexibilitet, hög förutsägelseprestanda samt låga tränings- och förutsägelsetider. Huvudsyftet med denna studie är på klassificerings- och regressiondataset utvärdera och belysa prestandaskillnaderna av 5 moderna samt 2 äldre GB-algoritmer. Målet är att avgöra vilken av dessa moderna algoritmer som presterar i genomsnitt bäst utifrån på flera utvärderingsmått. Initialt utfördes en teoretisk förstudie inom det aktuella forskningsområdet. Algoritmerna XGBoost, LightGBM, CatBoost, AdaBoost, SnapBoost, KTBoost, GBDT implementerades på plattformen Google Colab. Där utvärderades dess respektive, tränings- och förutsägelsestid samt prestandamåtten, uppdelat i ROCAUC och Log Loss för klassificering samt R2 och RMSE för regression. Resultaten visade att det generellt var små skillnader mellan dom olika testade algoritmerna. Med undantag för AdaBoost som i allmänhet, med större marginal, hade den sämsta prestandan. Därmed gick det inte i denna jämförelse utse en klar vinnare. Däremot presterade SnapBoost väldigt bra på flera utvärderingsmått. Modellresultaten är generellt sätt väldigt begränsade och bundna till det applicerade datasetet vilket gör att det överlag är väldigt svårt att generalisera det till andra datauppsättningar. Detta speglar sig från resultaten med svårigheten att identifiera ett ML-ramverk som utmärker sig och presterar bra i alla scenarier. / Machine learning (ML) is today a very relevent, popular and actively researched area. As a result, today there exits a large numer of different advanced and modern ML algorithms. The difficulty is to identify among these the most optimal to apply to one’s area of application. Algorithms based on Gradient Boosting (GB) have been shown to have a very wide range of application areas, flexibility, high prediction performance and low training and prediction times. The main purpose of this study is on classification and regression datasets evaluate and illustrate the performance differences of 5 modern and 2 older GB algorithms. The goal is to determine which of these modern algorithms, on average, performs best on the basis of several evaluation metrics. Initially, a theoretical feasibility study was carried out in the current research area. The algorithms XGBoost, LightGBM, CatBoost, AdaBoost, SnapBoost, KTBoost, GBDT were implemented on the Google Colab platform. There, respective training and prediction time as well as the performance metrics were evaluated, divided into ROC-AUC and Log Loss for classification and R2 and RMSE for regression. The results showed that there were generally small differences between the different algorithms tested. With the exception of AdaBoost which in general, by a larger margin, had the worst performance. Thus, it was not possible in this comparison to nominate a clear winner. However, SnapBoost performed very well in several evaluation metrics. The model results are generally very limited and bound to the applied dataset, which makes it generally very difficult to generalize it to other data sets. This is reflected in the results with the difficulty of identifying an ML framework that excels and performs well in all scenarios.

The issue of the relationship between water discharge and the suspended sediment loads is a globally highly addressed topic. Knowing the suspended sediment loads in the streams avoids problems with over-filling of water cannons and thus prevents insufficient capacity of water reservoirs. This thesis is partly a follow-up to the bachelor thesis, which extends and introduces new procedures. Neural networks, more specifically multilayer perceptron neural networks, are used to analyse the relationship between water discharge and suspended sediment loads. The results of the networks are then processed in Excel into graphs and evaluated using the coefficient of determination, Nash-Sutcliffe coefficient and RMSE coefficient. The practical application is solved on two profiles - the profile Podhradí nad Dyjí and the profile Židlochovice. Each profile is examined in a different period.

Econometric and predictive modeling techniques are two popular forecasting techniques. Both ofthese techniques have their own advantages and disadvantages. In this thesis some econometricmodels are considered and compared to predictive models using sales data for five products fromICA a Swedish retail wholesaler. The econometric models considered are regression model,exponential smoothing, and ARIMA model. The predictive models considered are artificialneural network (ANN) and ensemble of neural networks. Evaluation metrics used for thecomparison are: MAPE, WMAPE, MAE, RMSE, and linear correlation. The result of this thesisshows that artificial neural network is more accurate in forecasting sales of product. But it doesnot differ too much from linear regression in terms of accuracy. Therefore the linear regressionmodel which has the advantage of being comprehensible can be used as an alternative to artificialneural network. The results also show that the use of several metrics contribute in evaluatingmodels for forecasting sales. / Program: Magisterutbildning i informatik

In this project a steady-state building load for Constant Volume Terminal Reheat (CVTR), Dual Duct Constant Volume (DDCV) and Dual Duct Variable Air Volume (DDVAV) systems for the Zachry Engineering Building has been modeled. First, the thermal resistance values of the building structure have been calculated. After applying some assumptions, building characteristics were determined and building loads were calculated using the diversified loads calculation method. By having the daily data for six months for the Zachry building, the input to the CVTR, DDCV and DDVAV Microsoft Excel code were prepared for starting the simulation. The air handling units for the Zachry building are Dual Duct Variable Air Volume (DDVAV) systems. The calibration procedure has been used to compare the calibration signatures with characteristic signatures in order to determine which input variables need to be changed to achieve proper calibration. Calibration signatures are the difference between measured energy consumption and simulated energy consumption as a function of temperature. Characteristic signatures are the energy consumption as a function of temperature obtained by changing the value of input variables of the system. The base simulated model of the DDVAV system has been changed according to the characteristic signatures of the building and adjusted to get the closest result to the measured data. The simulation method for calibration could be used for energy audits, improving energy efficiency, and fault detection. In the base model of DDVAV, without any changes in the input, the chilled water consumption had an Root Mean Square Error (RMSE) of 56.705577 MMBtu/day and an Mean Bias Error (MBE) of 45.763256 MMBtu/day while hot water consumption had an RMSE of 1.9072574 MMBtu/day and an MBE of 45.763256 MMBtu/day. In the calibration process, system parameters such as zone temperature, cooling coil temperature, minimum supply air and minimum outdoor air have been changed. The decisions for varying the parameters were based on the characteristic signatures provided in the project. After applying changes to the system parameters, RMSE and MBE for both hot and cold water consumption were significantly reduced. After changes were applied, chilled water consumption had an RMSE of 12.749868 MMBtu/day and an MBE of 3.423188 MMBtu/day, and hot water consumption had an RMSE of 1.6790 MMBtu/day and an MBE 0.12513 of MMBtu/day.

MSc (Statistics) / Department of Statistics / The baseline parameter vector in a discrete-time survival model is determined by the number of time points. The larger the number of the time points, the higher the dimension of the baseline parameter vector which often leads to biased maximum likelihood estimates. One of the ways to overcome this problem is to use a simpler parametrization that contains fewer parameters. A simulation approach was used to compare the accuracy of three variants of penalised regression spline methods in smoothing the baseline hazard function. Root mean squared error (RMSE) analysis suggests that generally all the smoothing methods performed better than the model with a discrete baseline hazard function. No single smoothing method outperformed the other smoothing methods. These methods were also applied to data on age at rst alcohol intake in Thohoyandou. The results from real data application suggest that there were no signi cant di erences amongst the estimated models. Consumption of other drugs, having a parent who drinks, being a male and having been abused in life are associated with high chances of drinking alcohol very early in life. / NRF

This thesis wants to explore the forecasting potential of the multi-factor models to predict excess returns of the aggregated portfolio of the European stock mar- ket. These factors provided by Fama and French and Carhart are well-known in the field of asset pricing, we also add several financial and macroeconomic factors according to the literature. We establish a benchmark model of ARIMA and we compare the forecasting errors of OLS and the LSTM neural networks. Both models take the lagged excess returns and the inputs. We measure the performance with the root mean square error and mean absolute error. The results suggest that neural networks are in this particular task capable of bet- ter predictions given the same input as OLS but their forecasting error is not significantly lower according to the Diebold-Mariano test. JEL Classification C45, C53, C61, E37, G11, G15 Keywords Stocks, European market, Neural networks, LSTM, Factor Models, Fama-French, Predic- tions, RMSE Title Does LSTM neural network improve factor mod- els' predictions of the European stock market?

In this paper we conduct an out-of-sample forecasting exercise for monthly Swedish air traveler volumes. The models considered are multiplicative seasonal ARIMA, Neural network autoregression, Exponential smoothing, the Prophet model and a Random Walk as a benchmark model. We divide the out-of-sample data into three different evaluation periods: Pre-COVID-19, during COVID-19 and Post-COVID-19 for which we calculate the MAE, MAPE and RMSE for each model in each of these evaluation periods. The results show that for the Pre-COVID-19 period all models produce accurate forecasts, in comparison to the Random Walk model. For the period during COVID-19, no model outperforms the Random Walk, with only Exponential smoothing performing as well as the Random Walk. For the period Post-COVID-19, the best performing models are Random Walk, SARIMA and Exponential smoothing, with all aforementioned models having similar performance.

This study involves investigating the impacts of ignoring nested data structure in Rasch/1PL item response theory (IRT) model via a two-level and three-level hierarchical generalized linear model (HGLM). Currently, Rasch/IRT models are frequently used in educational and psychometric researches for data obtained from multistage cluster samplings, which are more likely to violate the assumption of independent observations of examinees required by Rasch/IRT models. The violation of the assumption of independent observation, however, is ignored in the current standard practices which apply the standard Rasch/IRT for the large scale testing data. A simulation study (Study Two) was conducted to address this issue of the effects of ignoring nested data structure in Rasch/IRT models under various conditions, following a simulation study (Study One) to compare the performances of three methods, such as Penalized Quasi-Likelihood (PQL), Laplace approximation, and Adaptive Gaussian Quadrature (AGQ), commonly used in HGLM in terms of accuracy and efficiency in estimating parameters. As expected, PQL tended to produce seriously biased item difficulty estimates and ability variance estimates whereas almost unbiased for Laplace or AGQ for both 2-level and 3-level analysis. As for the root mean squared errors (RMSE), three methods performed without substantive differences for item difficulty estimates and ability variance estimates in both 2-level and 3-level analysis, except for level-2 ability variance estimates in 3-level analysis. Generally, Laplace and AGQ performed similarly well in terms of bias and RMSE of parameter estimates; however, Laplace exhibited a much lower convergence rate than that of AGQ in 3-level analyses. The results from AGQ, which produced the most accurate and stable results among three computational methods, demonstrated that the theoretical standard errors (SE), i.e., asymptotic information-based SEs, were underestimated by at most 34% when 2-level analyses were used for the data generated from 3-level model, implying that the Type I error rate would be inflated when the nested data structures are ignored in Rasch/IRT models. The underestimated theoretical standard errors were substantively more severe as the true ability variance increased or the number of students within schools increased regardless of test length or the number of schools. / Ph. D.

Les méthodes et les outils de génération automatique ou semi-automatique de modèles 3D urbains se développent rapidement, mais l’évaluation de la qualité de ces modèles et des données spatiales sur lesquelles ils s’appuient n’est que rarement abordée. Notre objectif est de proposer une approche multidimensionnelle standard pour évaluer la qualité des modèles 3D de bâtiments en 1D, 2D et 3D. Deux méthodes sont présentées pour l'évaluation 1D. La première se base sur l’analyse de l’erreur moyenne quadratique en X, Y et Z. La deuxième solution s’appuie sur les instructions parues au Journal Officiel du 30 octobre 2003 et exigeant le respect de classes de précisions. L'approche que nous proposons se penche sur le calcul d'indices de qualité fréquemment rencontrés dans la littérature. L'originalité de notre approche réside dans le fait que les modèles employés en entrée ne se limitent pas au mode raster, mais s'étendent au mode vecteur. Il semble évident que les modèles définis en mode vecteur s'avèrent plus fidèles à la réalité qu'en mode raster. Les indices de qualité 2D et 3D calculés montrent que les modèles 3D de bâtiments extraits à partir des couples d’images stéréoscopiques sont cohérents. Les modèles reconstruits à partir du LiDAR sont moins exacts. En conclusion, cette thèse a abouti à l’élaboration d’une approche d’évaluation multidimensionnelle de bâtiments en 3D. L’approche proposée dans cette thèse est adaptée et opérationnelle pour des modèles vectoriels et rasters de bâtiments 3D simplifiés. / Methods and tools for automatic or semi-automatic generation of 3D city models are developing rapidly, but the quality assessment of these models and spatial data are rarely addressed. A comprehensive evaluation in 3D is not trivial. Our goal is to provide a standard multidimensional approach for assessing the quality of 3D models of buildings in 1D, 2D and 3D. Two methods are applied. The first one is done by computing Root Mean Square Errors (RMSE) based on the deviations between both models (reference and test), in X, Y and Z directions. Second method is performed by applying the French legal text (arrêté sur les classes de précision) that is based on the instructions published in the Official Journal from October 30, 2003. These indices pass through the space discretization in pixels or voxels for measuring the degree of superposition of 2D or 3D objects. The originality of this approach is built on the fact that the models used as input are not only limited to raster format, but also extended to vector format. The results of statistics of the quality indices calculated for assessing the building models show that the 3D building models extracted from stereo-pairs are close from each other. Also, the models reconstructed from LiDAR are less accurate than the models reconstructed from aerial images alone. In conclusion, the quality evaluation of 3D building models has been achieved by applying the proposed multi-dimensional approach. This approach is suitable for simplified 3D building vector models created from aerial images and/or LiDAR datasets.