The main aim of this master's thesis is to describe the subject of the implementation of decision boundaries with the help of artificial neural networks. The objective is to present theoretical knowledge concerning this field and on practical examples prove these statements. The work contains basic theoretical description of the field of pattern recognition and the field of feature based representation of objects. A classificator working on the basis of Bayes decision is presented in this part, and other types of classificators are named as well. The work then deals with artificial neural networks in more detail; it contains a theoretical description of their function and their abilities in the creation of decision boundaries in the feature plane. Examples are shown from literature for the use of neural networks in corresponding problems. As part of this work, the program ANN-DeBC was created using Matlab, for the generation of practical results about the usage of feed-forward neural networks for the implementation of decision boundaries. The work contains a detailed description of this program, and the achieved results are presented and analyzed. It is shown as well, how artificial neural networks are creating decision boundaries in the form of geometrical shapes. The effects of the chosen topology of the neural network and the number of training samples on the success of the classification are observed, and the minimal values of these parameters are determined for the successful creation of decision boundaries at the individual examples. Furthermore, it's presented how the neural networks behave at the classification of realistically distributed training samples, and what methods can affect the shape of the created decision boundaries.

The diploma thesis deal with a priority network switch whose model was made in programming environment Matlab - Simulink. Problem of optimal switching is solved by Hopfield’s artificial neural network. Produce of the diploma thesis is a model of packet switch and time-severity comparison of optimalization problem solved with or without artificial neural network. The thesis was developed in research project MSM 0021630529 Intelligent Systems in Automation.

This work deals with the usage of neural network for the purpose of stock market prediction. A basic stock market theory and trading approaches are mentioned at the beginning of this work. Then neural networks and their application are discussed with their deeper description. Similar approaches are referred and finally two new prediction systems are designed. These systems are utilized by proposed trading model and tested on selected data. The results are compared to human and random trading models and new development steps are devised at the end of this work.

This paper addresses the problem of detecting K-complexes in sleep EEG. The study of sleep has become very essential to diagnose the brain disorders and analysis of brain activities. Since Kcomplex can have a wide variety of shapes it is very difficult to detect the K-complexes manually. In this paper, I present an automatic method for K-complexes detection based wavelet transform,TKEO and method for classification using feedforward multilayer neural network designed in Matlab. Detection performance reached the value approx. from 52,9 to 83,6 %.

The thesis describe issue of segmentation and classification of Petri dishes colored images. There is proposed a segmentation method that extracts positions of cells from the image. Another techniques focues on classification to groups according to the parameters obtained from images. Reliability of each optimalized algorithm is tested on database containing 250 colored images received from company BioVendor Instruents a.s.

In this thesis, methods of implementing an online gesture recognition system for the novel Stylaero Board device are investigated. Two methods are evaluated - one based on LSTMs and one based on HMMs - on three kinds of gestures: Tap, circle, and flick motions. A method’s performance was measured in its accuracy in determining both whether any of the above listed gestures were performed and, if so, which gesture, in an online single-pass scenario. Insight was acquired regarding the technical challenges and possible solutions to the online aspect of the problem. Poor performance was, however, observed in both methods, with a likely culprit identified as low quality of training data, due to an arduous and complex gesture performance capturing process. Further research improving on the process of gathering data is suggested.

<div>Currently, people in North America spend roughly 90% of their time indoors. Therefore, it is important to create comfortable, healthy, and productive indoor environments for the occupants. Unfortunately, our resulting indoor environments are still very poor, especially in multi-occupant rooms. In addition, energy consumption in residential and commercial buildings by HVAC systems and lighting accounts for about 41% of primary energy use in the US. However, the current methods for simulating building energy consumption are often not accurate, and various types of occupant behavior may explain this inaccuracy.</div><div>This study first developed artificial neural network models for predicting thermal comfort and occupant behavior in indoor environments. The models were trained by data on indoor environmental parameters, thermal sensations, and occupant behavior collected in ten offices and ten houses/apartments. The models were able to predict similar acceptable air temperature ranges in offices, from 20.6 °C to 25 °C in winter and from 20.6 °C to 25.6 °C in summer. We also found that the comfortable air temperature in the residences was 1.7 °C lower than that in the offices in winter, and 1.7 °C higher in summer. The reason for this difference may be that the occupants of the houses/apartments were responsible for paying their energy bills. The comfort zone obtained by the ANN model using thermal sensations in the ten offices was narrower than the comfort zone in ASHRAE Standard 55, but that using behaviors was wider.</div><div>Then this study used the EnergyPlus program to simulate the energy consumption of HVAC systems in office buildings. Measured energy data were used to validate the simulated results. When using the collected behavior from the offices, the difference between the simulated results and the measured data was less than 13%. When a behavioral ANN model was implemented in the energy simulation, the simulation performed similarly. However, energy simulation using constant thermostat set point without considering occupant behavior was not accurate. Further simulations demonstrated that adjusting the thermostat set point and the clothing could lead to a 25% variation in energy use in interior offices and 15% in exterior offices. Finally, energy consumption could be reduced by 30% with thermostat setback control and 70% with occupancy control.</div><div>Because of many contextual factors, most previous studies have built data-driven behavior models with limited scalability and generalization capability. This investigation built a policy-based reinforcement learning (RL) model for the behavior of adjusting the thermostat and clothing level. We used Q-learning to train the model and validated with collected data. After training, the model predicted the behavior with R2 from 0.75 to 0.80 in an office building. This study also transferred the behavior knowledge of the RL model to other office buildings with different HVAC control systems. The transfer learning model predicted with R2 from 0.73 to 0.80. Going from office buildings to residential buildings, the transfer learning model also had an R2 over 0.60. Therefore, the RL model combined with transfer learning was able to predict the building occupant behavior accurately with good scalability, and without the need for data collection.<br></div><div><div>Unsuitable thermostat settings lead to energy waste and an undesirable indoor environment, especially in multi-occupant rooms. This study aimed to develop an HVAC control strategy in multi-occupant offices using physiological parameters measured by wristbands. We used an ANN model to predict thermal sensation from air temperature, relative humidity, clothing level, wrist skin temperature, skin relative humidity and heart rate. Next, we developed a control strategy to improve the thermal comfort of all the occupants in the room. The control system was smart and could adjust the thermostat set point automatically in real time. We improved the occupants’ thermal comfort level that over half of the occupants reported feeling neutral, and fewer than 5% still felt uncomfortable. After coupling with occupancy-based control by means of lighting sensors or wristband Bluetooth, the heating and cooling loads were reduced by 90% and 30%, respectively. Therefore, the smart HVAC control system can effectively control the indoor environment for thermal comfort and energy saving.</div><div>As for proposed studies in the future, at first, we will use more advanced sensors to collect more kinds of occupant behavior-related data. We will expand the research on more occupant behavior related to indoor air quality, noise and illuminance level. We can use these data to recognize behavior instead of questionnaire survey now. We will also develop a personalized zonal control system for the multi-occupant office. We can find the number and location of inlet diffusers by using inverse design.</div></div>

This study investigates the predictive performance of two different machine learning (ML) models on the stock market and compare the results. The chosen models are based on artificial neural networks (ANN) and random forests (RF). The models are trained on two separate data sets and the predictions are made on the next day closing price. The input vectors of the models consist of 6 different financial indicators which are based on the closing prices of the past 5, 10 and 20 days. The performance evaluation are done by analyzing and comparing such values as the root mean squared error (RMSE) and mean average percentage error (MAPE) for the test period. Specific behavior in subsets of the test period is also analyzed to evaluate consistency of the models. The results showed that the ANN model performed better than the RF model as it throughout the test period had lower errors compared to the actual prices and thus overall made more accurate predictions. / Denna studie undersöker hur väl två olika modeller inom maskininlärning (ML) kan förutspå aktiemarknaden och jämför sedan resultaten av dessa. De valda modellerna baseras på artificiella neurala nätverk (ANN) samt random forests (RF). Modellerna tränas upp med två separata datamängder och prognoserna sker på nästföljande dags stängningskurs. Indatan för modellerna består av 6 olika finansiella nyckeltal som är baserade på stängningskursen för de senaste 5, 10 och 20 dagarna. Prestandan utvärderas genom att analysera och jämföra värden som root mean squared error (RMSE) samt mean average percentage error (MAPE) för testperioden. Även specifika trender i delmängder av testperioden undersöks för att utvärdera följdriktigheten av modellerna. Resultaten visade att ANN-modellen presterade bättre än RF-modellen då den sett över hela testperioden visade mindre fel jämfört med de faktiska värdena och gjorde därmed mer träffsäkra prognoser.

In recent years, the Buy Now Pay Later (BNPL) consumer credit industry associated with e-commerce has been rapidly emerging as an alternative to credit cards and traditional consumer credit products. In parallel, the regulation IFRS 9 was introduced in 2018 requiring creditors to become more proactive in forecasting their Expected Credit Losses and include the impact of macroeconomic factors. This study evaluates several methods of supervised statistical learning to model the Probability of Default (PD) for BNPL credit contracts. Furthermore, the study analyzes to what extent macroeconomic factors impact the prediction under the requirements in IFRS 9 and was carried out as a case study with the Swedish fintech firm Klarna. The results suggest that XGBoost produces the highest predictive power measured in Precision-Recall and ROC Area Under Curve, with ROC values between 0.80 and 0.91 in three modeled scenarios. Moreover, the inclusion of macroeconomic variables generally improves the Precision-Recall Area Under Curve. Real GDP growth, housing prices, and unemployment rate are frequently among the most important macroeconomic factors. The findings are in line with previous research on similar industries and contribute to the literature on PD modeling in the BNPL industry, where limited previous research was identified. / De senaste åren har Buy Now Pay Later (BNPL) snabbt vuxit fram som ett alternativ till kreditkort och traditionella kreditprodukter, i synnerhet inom e-handel. Dessutom introducerades 2018 det nya regelverket IFRS 9, vilket kräver att banker och andra kreditgivare ska bli mer framåtblickande i modelleringen av sina förväntade kreditförluster, samt ta hänsyn till effekter från makroekonomiska faktorer. I denna studie utvärderas flera metoder inom statistisk inlärning för att modellera Probability of Default (PD), sannolikheten att en kreditförlust inträffar, för BNPL-kreditkontrakt. Dessutom analyseras i vilken utsträckning makroekonomiska faktorer påverkar modellernas prediktiva förmågor enligt kraven i IFRS 9. Studien genomfördes som en fallstudie med det svenska fintechföretaget Klarna. Resultaten tyder på att XGBoost har den största prediktionsförmågan mätt i Precision-Recall och ROC Area Under Curve, med ROC-värden mellan 0.80 och 0.91 i tre scenarier. Inkludering av makroekonomiska variabler förbättrar generellt PR-Area Under Curve. Real BNP-tillväxt, bostadspriser och arbetslöshet återfinns frekvent bland de viktigaste makroekonomiska faktorerna. Resultaten är i linje med tidigare forskning inom liknande branscher och bidrar till litteraturen om att modellera PD i BNPL-branschen där begränsad tidigare forskning hittades.

Comprehensive information on traffic flow is essential for vehicular emission monitoring and traffic control. However, such information is not observable everywhere and anytime on the road because of high installation costs and malfunctions of stationary sensors. In order to compensate for stationary sensors’ weakness, this thesis analyses an approach for inferring traffic flows from mobile data provided by INRIX, a commercial crowd-sourced traffic dataset with wide spatial coverage and high quality. The idea is to develop Artificial Neural Network (ANN)-based models to automatically extract relations between traffic flow and INRIX measurements, e.g., speed and travel time, from historical data considering temporal and spatial dependencies. We conducted experiments using four weeks of data from INRIX and stationary sensors on two adjacent road segments on the E4 highway in Stockholm. Models are validated via traffic flow estimation based on one week of INRIX data. Compared with the traditional approach that fits the stationary flow-speed relationship based on the multi-regime model, the new approach greatly improves the estimation accuracy. Moreover, the results indicate that the new approach’s models have better resistance to the drift of input variables and can decrease the deterioration of estimation accuracy on the road segment without a stationary sensor. Hence, the new approach may be more appropriate for estimating traffic flows on the nearby road segments of a stationary sensor. The approach provides a highly automated means to build models adaptive to datasets and improves estimation and imputation accuracy. It can also easily integrate new data sources to improve the models. Therefore, it is very suitable to be applied to Intelligent Transport Systems (ITS) for traffic monitor and control in the context of the Internet of Things (IoT) and Big Data. / Information om trafikflödet är nödvändig för övervakning av fordonsutsläpp och trafikstyrning. Trafikflöden kan dock inte observeras överallt och när som helst på vägen på grund av höga installationskostnader och t.ex. funktionsstörningar hos stationära sensorer. För att kompensera för stationära sensorers svagheter analyseras i detta arbete ett tillvägagångssätt för att estimera trafikflöden från mobila data som tillhandahålls av INRIX. Detta kommersiella dataset innehåller restider som kommer från användare av bl.a. färdnavigatorer i fordon och som har en bred rumslig täckning och hög kvalitet. Idén är att utveckla modeller baserade på artificiellt neuronnät för att automatiskt extrahera samband mellan trafikflödesdata och restidsdata från INRIX-mätningarna baserat på historiska data och med hänsyn till tidsmässiga och rumsliga beroenden. Vi utförde experiment med fyra veckors data från INRIX och från stationära sensorer på två intilliggande vägsegment på E4:an i Stockholm. Modellerna valideras med hjälp av estimering av trafikflöde baserat på en veckas INRIX- data. Jämfört med det traditionella tillvägagångssättet som anpassar stationära samband mellan trafikflöde och hastighet baserat på fundamentaldiagram, förbättrar det nya tillvägagångssättet noggrannheten avsevärt. Dessutom visar resultaten att modellerna i den nya metoden bättre hanterar avvikelser i ingående variabler och kan öka noggrannheten på estimatet för vägsegmentet utan stationär sensor. Den nya metoden kan därför vara lämplig för att uppskatta trafikflöden på vägsegment närliggande en stationär sensor. Metodiken ger ett automatiserat sätt att bygga modeller som är anpassade till datamängderna och som förbättrar noggrannheten vid estimering av trafikflöden. Den kan också enkelt integrera nya datakällor. Metodiken är lämplig att tillämpa på tillämpningar inom intelligenta transportsystem för trafikövervakning och trafikstyrning.