The need for a robust model for predicting the value of condominiums and houses are becoming more apparent as further evidence of systematic errors in existing models are presented. Traditional valuation methods fail to produce good predictions of condominium sales prices and systematic patterns in the errors linked to for example the repeat sales methodology and the hedonic pricing model have been pointed out by papers referenced in this thesis. This inability can lead to monetary problems for individuals and in worst-case economic crises for whole societies. In this master thesis paper we present how a predictive model constructed from a multilayer perceptron can predict the price of a condominium in the centre of Stockholm using objective data from sources publicly available. The value produced by the model is enriched with a predictive interval using the Inductive Conformal Prediction algorithm to give a clear view of the quality of the prediction. In addition, the Multilayer Perceptron is compared with the commonly used Support Vector Regression algorithm to underline the hallmark of neural networks handling of a broad spectrum of features. The features used to construct the Multilayer Perceptron model are gathered from multiple “Open Data” sources and includes data as: 5,990 apartment sales prices from 2011- 2013, interest rates for condominium loans from two major banks, national election results from 2010, geographic information and nineteen local features. Several well-known techniques of improving performance of Multilayer Perceptrons are applied and evaluated. A Genetic Algorithm is deployed to facilitate the process of determine appropriate parameters used by the backpropagation algorithm. Finally, we conclude that the model created as a Multilayer Perceptron using backpropagation can produce good predictions and outperforms the results from the Support Vector Regression models and the studies in the referenced papers. / Behovet av en robust modell för att förutsäga värdet på bostadsrättslägenheter och hus blir allt mer uppenbart alt eftersom ytterligare bevis på systematiska fel i befintliga modeller läggs fram. I artiklar refererade i denna avhandling påvisas systematiska fel i de estimat som görs av metoder som bygger på priser från repetitiv försäljning och hedoniska prismodeller. Detta tillkortakommandet kan leda till monetära problem för individer och i värsta fall ekonomisk kris för hela samhällen. I detta examensarbete påvisar vi att en prediktiv modell konstruerad utifrån en “Multilayer Perceptron” kan estimera priset på en bostadsrättslägenhet i centrala Stockholm baserad på allmänt tillgängligt data (“Öppen Data”). Modellens resultat har utökats med ett prediktivt intervall beräknat utifrån “Inductive Conformal Prediction”- algoritmen som ger en klar bild över estimatets tillförlitlighet. Utöver detta jämförs “Multilayer Perceptron”-algoritmen med en annan vanlig algoritm för maskinlärande, den så kallade “Support Vector Regression” för att påvisa neurala nätverks kvalité och förmåga att hantera dataset med många variabler. De variabler som används för att konstruera “Multilayer Perceptron”-modellen är sammanställda utifrån allmänt tillgängliga öppna datakällor och innehåller information så som: priser från 5990 sålda lägenheter under perioden 2011- 2013, ränteläget för bostadsrättslån från två av de stora bankerna, valresultat från riksdagsvalet 2010, geografisk information och nitton lokala särdrag. Ett flertal välkända förbättringar för “Multilayer Perceptron”-algoritmen har applicerats och evaluerats. En genetisk algoritm har använts för att stödja processen att hitta lämpliga parametrar till “Backpropagation”-algoritmen. I detta arbete drar vi slutsatsen att modellen kan producera goda förutsägelser med en modell konstruerad utifrån ett neuralt nätverk av typen “Multilayer Perceptron” beräknad med “backpropagation”, och därmed utklassar de resultat som levereras av Support Vector Regression modellen och de studier som refererats i denna avhandling

1 ABSTRACT Background Breast cancer cell population characteristics are used in common clinical practice for estimation of prognosis of the malignant disease (prognostic factors) and for prediction of reactivity of the tumor to certain therapeutic modality (predictive factors). Also axillary lymph node status is an independent prognostic factor in women with early breast cancer. Therefore, surgical excision and following histopathological examination of the nodes is the obligatory part of primary breast cancer surgery. The extension of axillary surgery varies widely, although sentinel lymph node biopsy is considered to be the standard procedure. However, it must be admitted that this type of procedure need not be optimal for all the breast cancer patients. Aims of the study The aim of this study is the verify the hypothesis whether or not the axillary lymph node metastatic affection can be effectively estimated using non-surgical methods - i.e. by evaluation of the combination of prognostic and predictive factors of the primary breast tumor. Statistical model composed on the basis of data of early breast cancer patients is the basic tool for this prediction. Application of this model In everyday practice can enable to adjust the extent of axillary surgery for each individual patient. Patients and methods A...

El presente trabajo muestra la situación actual de la demanda de Gas Liquado de Petroleo (GLP) en el mercado peruano con respecto al parque automotor durante los últimos 6 años. El objetivo general es predecir la demanda de GLP para el segundo semestre del año 2021, a través de las variables más relevantes a fin de conocer si la producción local más la importación de este tipo de combustible (GLP) será la suficiente para cubrir la demanda del sector automotriz. La metodología utilizada por el equipo de ciencia de datos es Cross Industry Standard Process for Data Mining (CRISP-DM), la cual consiste en seguir una serie de diez etapas, en cada una de ellas se ira descubriendo y analizando las variables que serán relevantes para la elaboración del modelo deseado. El modelo seleccionado por el equipo de ciencia de datos es el modelo de aprendizaje predictivo ya que este agrupa varias técnicas estadísticas de modelización, lo cual incluye algoritmos de aprendizaje automático. Posteriormente las Herramientas que se utilizarán para un mejor Análisis y entendimiento de la problemática serán Power BI, KNime y Python. / This paper shows the current situation of Liquefied Petroleum Gas (LPG) demand in the Peruvian market with respect to the vehicle fleet during the last 6 years. The general objective is to predict the LPG demand for the second semester of the year 2021, through the most relevant variables to know if the local production plus the import of this type of fuel (LPG) will be enough to cover the demand of the automotive sector. The methodology used by the data science team is Cross Industry Standard Process for Data Mining (CRISP-DM), which consists of following a series of ten stages, in each of which the variables that will be relevant for the elaboration of the desired model will be discovered and analyzed. The model selected by the data science team is the predictive learning model because it groups several statistical modeling techniques, including machine learning algorithms. Subsequently, the tools to be used for a better analysis and understanding of the problem will be Power BI, KNime and Python. / Trabajo de investigación

In the nearest future, the human driver is viewed as a reliable backup even for the fully automated road motor vehicles (cars). Indeed, the driver is assumed to swiftly take the control of the car in cases of suddenly occurring (i) challenging environmental conditions, (ii) complex unforeseen driving situations, or (iii) degradation of performance of the car. However, due to the cognitive overload in such a sudden, stressful takeover of the control, the driver would often experience the startle effect, which usually results in an unconscious, instinctive, yet incorrect response. An extreme case of startle is freezing, in which the driver might be incapable to respond to the sudden takeover of control at all. The possible approaches to alleviate the startle during the takeover of control (i.e., the automation startle) include an offset- (i.e., either early- or delayed-), gradual yielding the controls to the driver. In the cases considered above, however, these approaches are hardly applicable because of (i) the presumed unpredictability of the events that result in the need of takeover of control, and (ii) the severe time constraints of the latter. Conversely, the objective of our research is to propose an approach of minimizing the need of yielding the control to the driver in challenging environmental conditions by guaranteeing an adequate automated control in these conditions. Focusing on slippery roads as an instance of challenging conditions, and steering control as an instance of control, we aim at developing such an automated steering that controls the car adequately in various road surfaces featuring low friction coefficients without the need of driver’s intervention.In order to develop such an automated steering we employed an in-house evolutionary computation framework – XML-based genetic programming (XGP) – which offers a flexible, portable, and human readable representation of the evolved optimal steering functions. The trial runs of the evolved steering functions were performed in the Open Source Racing Car Simulator (TORCS), which features a realistic, yet computationally efficient simulation of the car and its environment. The obtained experimental results indicate that due to the challenging dynamics of the unstable car on slippery roads, neither the canonical (tuned) servo-control (as a variant of PD) nor the (tuned) PID-controller could control the car adequately on slippery roads. On the other hand, the controller, featuring a relaxed, arbitrary structure evolved by XGP outperforms both the servo- and PID controllers in that it results in a minimal deviation of the car from its intended trajectory in rainy, snowy, and icy road conditions. Moreover, the evolved steering that employs anticipated perceptions is even superior as it could anticipate the imminent understeering of the car at the entry of the turns and consequently – to compensate for such an understeering by proactively turning the steering wheels in advance – well before entering the turn. The obtained results suggest a human competitiveness of the evolved automated steering as it outperforms the commonly used alternative steering controllers proposed by human experts. The research could be viewed as a step towards the evolutionary development of automated steering of cars in challenging environmental conditions. / 博士(工学) / Doctor of Philosophy in Engineering / 同志社大学 / Doshisha University

In this work we use a graph-theoretic representation of secondary RNA structure found in the database RAG: RNA-As-Graphs. We model the bonding of two RNA secondary structures to form a larger structure with a graph operation called merge. The resulting data from each tree merge operation is summarized and represented by a vector. We use these vectors as input values for a neural network and train the network to recognize a tree as RNA-like or not based on the merge data vector. The network correctly assigned a high probability of RNA-likeness to trees identified as RNA-like in the RAG database, and a low probability of RNA-likeness to those classified as not RNA-like in the RAG database. We then used the neural network to predict the RNA-likeness of all the trees of order 9. The use of a graph operation to theoretically describe the bonding of secondary RNA is novel.

The transportation industry stands for a big chunk of the worlds total carbon emissions. To counter this problem electric vehicles are seen as a good solution. However, these vehicles come at a greater cost and do not offer the same range as their less environmentally friendly counterpart. To lessen costs and development time when optimizing electric vehicles, simulations of the vehicles functionality can be utilized. One way of getting such simulations is to design a digital twin of the physical system. A digital twin is able to mimic the functionality of the physical system and can therefore offer well based indications of how a change in design will change the performance in reality. In this thesis a digital twin of the cooling system of an electric boat is designed with realistic results. Cooling systems in the scope of electric vehicles are of grave importance since the electric driveline becomes hot during use which can hinder performance of the vehicle. This is especially true for the high voltage batteries that tend to have quite a narrow range of temperatures within which performance is optimal. This thesis handles an attempt at optimizing the cooling system, replicated by the digital twin, by the use of a temperature predictive model. Three different machine learning models were tested and the resulting best model achieved a mean absolute error of 2.4 and a mean average percentage error of 5.7. However, the model was unable to foresee sudden temperature spikes and drops. A possible fix, that could not be tested in this thesis, would be to implement further input data such as driver profiles and/or GPS data with speed limits. / Transportindustrin står för en stor del av världens totala koldioxidutsläpp. För att motverka detta problem ses elfordon som en bra lösning. Dessa fordon kommer dock till en högre kostnad och erbjuder inte samma räckvidd som deras mindre miljövänliga motpart. För att minska kostnader och utvecklingstid vid optimering av elfordon kan simuleringar av fordonens funktionalitet användas. Ett sätt att få sådana simuleringar är att designa en digital tvilling av det fysiska systemet. En digital tvilling kan efterlikna det fysiska systemets funktionalitet och kan därför erbjuda välbaserade indikationer på hur en förändring i design kommer att förändra prestandan. I detta examensarbete designas en digital tvilling av kylsystemet i en elbåt med realistiska resultat. Kylsystem i elfordon är av stor betydelse eftersom den elektriska drivlinan blir varm under användning, vilket kan hindra fordonets prestanda. Detta gäller särskilt för högspänningsbatterierna som tenderar att ha ett ganska smalt temperaturintervall för optimal prestanda. Denna avhandling behandlar ett försök att optimera kylsystemet, replikerat av den digitala tvillingen, genom att använda en temperaturförutsende modell. Tre olika maskininlärningsmodeller testades och den resulterande bästa modellen uppnådde ett genomsnittligt absolut fel på 2.4 och ett genomsnittligt procentuellt fel på 5.7. Modellen kunde dock inte förutse plötsliga temperaturspikar och -fall. En möjlig fix, som inte kunde testas i denna avhandling, skulle vara att implementera ytterligare indata såsom förarprofiler och/eller GPS-data med hastighetsbegränsningar.

Le diagnostic de mort cérébrale est une étape complexe et chronophage lors de l'évaluation des patients en soins intensifs soupçonnés d'être en décès neurologique. Bien que les critères neurologiques cliniques qui déterminent la mort cérébrale soient largement acceptés dans le monde, le diagnostic reste imparfait et l'utilisation de tests auxiliaires tels que la perfusion tomographique cérébrale (CTP) est souvent nécessaire pour le confirmer. L'objectif principal de ce travail était d'explorer la faisabilité de classer la mort cérébrale à partir de scans CTP par le traitement automatique de l’image. Les scans CTP de l'étude prospective canadienne multicentrique de validation du CTP pour le diagnostic de décès neurologique ont été regroupées à partir de 11 sites participants (INDex-CTP, ClinicalTrials.gov, NCT03098511). Des caractéristiques spatiales et temporelles ont été extraites en utilisant une combinaison de deux modules de convolution et utilisées pour prédire la mort neurologique. Les performances du modèle ont également été évaluées sur différentes catégories de blessures cérébrales. Les études de 217 patients ont été utilisées pour entraîner le modèle. Nous rapportons une AUC de 0,79 (IC95 % 0,76-0,82), un score F1 de 0,82 (IC95 % 0,80-0,83), une précision de 0,92 (IC95 % 0,91-0,93), un rappel de 0,76 (CI95 % 0,72-0,79) ainsi qu'une valeur prédictive négative de 0,49 (CI95 % 0,45-0,53). En raison de la petite taille d'échantillon, nous n'avons pas effectué de tests statistiques sur des sous-ensembles de lésions cérébrales, mais avons signalé une valeur prédictive négative du modèle présumé plus élevée sur des blessures cérébrales anoxiques avec 0,82 (CI95 % 0,77-0,87). Ce modèle montre des preuves préliminaires soutenant la faisabilité de développer un réseau neuronal profond pour classer les patients comateux comme étant neurologiquement décédés ou non. Des recherches supplémentaires sont nécessaires pour valider et améliorer le modèle en utilisant des ensembles de données plus vastes et diversifiés. / The diagnostic of brain death is a complex and chronophage step when evaluating patients in critical care suspected of being neurologically deceased. Although the clinical neurological criteria that determine brain death are mostly accepted worldwide, the diagnosis remains imperfect and often the use of ancillary tests such as brain computed tomography perfusion (CTP) are required to confirm it. The main objective of this work was to explore the feasibility of classifying brain death from CTP scans using deep learning. CTP studies from a multicenter prospective diagnostic cohort study with the primary objective of evaluating the diagnostic accuracy of neurological death using CTP were pooled from 11 participating sites (INDex-CTP, ClinicalTrials.gov, NCT03098511). Spatial and temporal features were extracted using a combination of two convolution modules and used to predict neurological death. The performance of the model was also evaluated on subsets of cerebral injuries. 217 patients' studies were used to train the model. We report an AUC of 0.79 (IC95% 0.76-0.82), a F1 score of 0.82 (IC95% 0.80-0.83), a precision of 0.92 (IC95% 0.91-0.93), a recall of 0.76 (CI95% 0.72-0.79) as well as a negative predictive value of 0.49 (CI95% 0.45-0.53). Due to a lack of sample size, we did not perform statistical tests on subsets of cerebral injury, but report suspected higher model negative predictive value on anoxic cerebral injury with 0.82 (CI95% 0.77-0.87). This model shows preliminary evidence supporting the feasibility of developing a deep neural network to classify comatose patients as neurologically deceased or not. Additional research is needed to validate and refine the model by employing larger and more diverse datasets.

No description available.

No description available.

No description available.