|
Nowcasting the Real GDP Growth of the European Economies based on Machine Learning
Baylan, Su Hazal ; Kočenda, Evžen (advisor) ; Baruník, Jozef (referee)
This thesis analyzes the nowcasting of quarterly GDP growth for nine European economies using a dynamic factor model and four different machine learning models. These machine learning models are as follows: Ridge, Lasso, Elastic Net, and Random Forest. The data includes ten hard and fifteen soft indicators for each country in order to calculate GDP for each nowcasting iteration for pre-covid and covid periods. For machine learning, models are fed with the extracted factors that are obtained from the dynamic factor model, and for all nowcasting models expanding window approach is selected to estimate nowcasting iterations. The empirical finding indicates that overall machine learning models provide better forecasting accuracy compared to dynamic factor models and benchmark models for more stable periods, such as the period before Covid-19. On the other hand, for more volatile periods where the uncertainties are higher in economies, the dynamic factor model outperforms machine learning models in order to nowcast GDP growth. In addition to this, Random Forest is able to outperform all the alternative models for small economies such as Slovenia and Portugal for stable periods. JEL Classification C01, C33, C53, C83, E37 Keywords Nowcasting, DFM, Ridge, Lasso, Elastic Net, Random Forest Title Nowcasting...
|
|
Automatic diagnosis of the 12-lead ECG using deep learning
Blaude, Ondřej ; Chmelík, Jiří (referee) ; Provazník, Valentine (advisor)
The aim of this diploma thesis is to investigate the problematics of automatic ECG diagnostics, namely on twelve-lead recordings. This problem is solved by standard methods such as random forest, artificial neural networks or K-nearest neighbors. However, thanks to its ability to independently extract symptoms, deep learning methods are also popular. All these methods are described in the theoretical part. In the practical part, deep learning models were designed, functionality support was verified using data from the PhysioNet database. Two pilot models were created and subsequently optimized. From the entire parameter optimization procedure, three models are available, of which the best accuracy achieves an F1 score of 87.35% and 83.7%, and the second best achieves an F1 score of 77.74% and an accuracy of 84.53%. The results achieved are discussed and compared with those of similar publications.
|
|
Time Series Analysis
Budai, Samuel ; Bartík, Vladimír (referee) ; Burgetová, Ivana (advisor)
This thesis deals with the issue of time series analysis and its use in the detection of anomalies in industrial networks. AR-X, ARIMA, SARIMA, Random Forest, Facebook Prophet and XGB Boost algorithms were used in the solution to create prediction models. In addition, the work includes the implementation of an algorithm for detecting anomalies from prediction models as well as solving the problem of high seasonal period in the case of the SARIMA algorithm. Through the conducted research, it was found that with the use of selected algorithms, it is possible to predict industrial traffic for the purpose of detection, within which up to 90% of attacks were detected. The work also provides a solution to a high seasonal period using partial time series. These results allow the experimental integration of prediction-based detection into real industrial networks.
|
|
Testing possibilities to extract selected landscape characteristics for description of indication-relevant bird species habitats in the Krkonoše Mts. from remote sensing data
Polák, Mojmír ; Kupková, Lucie (advisor) ; Janík, Tomáš (referee)
The thesis uses remote sensing data from two spatial scales (Sentinel-2 with a 10 x 10 m pixel and PlanetScope 3 x 3 m. It explores the possibilities of extracting selected landscape characteristics (spectral indices, land cover type, landscape metrics). In order to evaluate which characteristics and at what scale are statistically significant for the occurrence of 23 selected bird species, species richness in quadrats and the number of species of the order Passeriformes in the Krkonoše Mountains. Data on species occurrence were mapped in the year 2012-2014 The strength of the relationship between characteristics and abundance was determined by Pearson's correlation coefficient. It could not be confirmed that data with higher spatial resolution would be more beneficial for extracting landscape characteristics. Overall, the landscape characteristics did not prove functional relationships for all selected species, but for some species, species richness, and order of Passeriformes, the assumption of elevation and land cover as important factors was confirmed. Land cover was analysed using the Random Forest supervised classification method in Google Earth Engine with an overall accuracy of 78 % for Sentinel-2 data, both in tundra and in the rest of the area, and 77 % for PlanetScoce data in tundra, 66...
|
|
Spatial Analysis of Czech Parliamentary Election: Comparison of Spatial Econometrics and Machine Learning
Černý, Jakub ; Šťastná, Lenka (advisor) ; Gregor, Martin (referee)
This thesis analyses the results of the Czech Parliamentary election in 2021 and attempts to explain the voting support of major political subjects by using aggregate data from Czech municipalities. Since the data evince spatial autocorrelation, it is necessary to specify a suitable spatial model. The thesis provides both empirical and economic evidence for the Spatial Durbin Error Model, which enables distinguishing the direct and indirect effects of particular independent variables and accounts for the spatial dependence of error terms. This method shows that variables describing the socio-economic characteristics of inhabitants, such as the share of entrepreneurs or people with university education, play the most significant role in explaining voting results and evince mostly the direct effects. On the contrary, variables describing municipalities, such as public spending or infrastructure, are more likely to impact the election result indirectly. Subsequently, the analysis is replicated using two tree-based machine learning algorithms and all models are evaluated based on their ability to predict the election results from unseen data. Even though machine learning methods estimate only relative variable importance instead of standard coefficients, this approach represents a perspective...
|
|
Detection of modern Slow DoS attacks
Jurek, Michael ; Jonák, Martin (referee) ; Sikora, Marek (advisor)
S rozvojem propojených zařízení v síti internet se počet útoků zvětšuje. Útočníci můžou zneužít takového zranitelného zařízení a vytvořit (D)DoS útok proti své oběti. Tyto útoky se stávají čím dál tím víc sofistikovanější. Proto byla vytvořena nová kategorie DoS útoků s názvem Pomalé DoS útoky, u kterých se útočník snaží napodobit chování standardního uživatele. Útočník se snaží využít všech možností, které mu transportní či aplikační protokol umožňují jako např. náhodné zahazování paketů, neodesílání nebo pozdržování zpráv. Na druhou stranu tvorba vlastních aplikačních výplní těchto protokolů může způsobit stav odepření služby na cíleném aplikačním serveru. Tato práce navrhuje klasifikaci síťových toků a volbu parametrů, které můžou pomoci s detekcí pomalých DoS útoků. Mezi vybranými pomalými DoS útoky jsou Slow Read, Slow Drop a Slow Next. Pro každý útok je popsán proces komunikace z pohledu transportní a aplikační vrstvy. Dále jsou vybrány důležité parametry popisující tyto útoky a v neposlední řadě jsou diskutovány metody a nástroje umožňující tvorbu takových útoků. Tato práce se zabývá možnostmi a nástroji tvorby spojení pro útok a diskutuje základní komunikační koncepty tvorby paralelních spojení. Dále je navržen vlastní generátor pomalých DoS útoků s velkým množstvím parametrů, pomocí nichž může útočník definovat vlastní pomalé DoS útoky. Následující část popisuje testovací prostředí pro testování generovaných útoků, scénáře a nástroje zachycování síťového provozu pro tvorbu vlastního datového souboru, jež je dále použit pro detekci pomalých DoS útoků pomocí metod strojového účení s učitelem. Konrétně jsou použity rozhodovací stromy a náhodné lesy k výběrů důležitých paramterů či sloupců použitelných pro detekci pomalých DoS útoků.
|
|
Mapping relict arctic-alpine tundra vegetation from multitemporal LiDAR data
Šrollerů, Alex ; Potůčková, Markéta (advisor) ; Lysák, Jakub (referee)
The thesis focuses on metrics of vertical structure of vegetation derived from UAV LiDAR data and their use for multitemporal classification of selected species of arctic-alpine tundra in the Krkonoše Mountains. The metrics are selected based on a literature search focusing on low and shrubby stands. Random Forest algorithm and permutation feature importance, drop column importance and individual predictor performance is used to determine the suitability of metrics for distinguishing tundra vegetation. Subsequently, a fusion with multispectral data is performed and influence of the LiDAR derived variables on the refinement of classification results is determined. The use of metrics derived from a digital surface model obtained by image correlation of multispectral data is also examined. Maximum height followed by minimum height, canopy relief ratio and coefficient of variation yielded the best results, they achieved an overall classification accuracy of 67.3% for Bílá louka meadow and 62.3% for Úpské rašeliniště bog. Fusion with multispectral data led to an increase in overall accuracy up to 2 %. In case of vegetation structure derived from the digital surface model, similar results were achieved apart from higher stands. LiDAR data did not prove to be beneficial in distinguishing grass communities...
|
|
Landcover classification of selected parts of Ethiopia based on machine learning method
Valchářová, Daniela ; Štych, Přemysl (advisor) ; Nedbal, Václav (referee)
Diploma thesis deals with the land cover classification in Sidama region of Ethiopia and 2 kebeles, Chancho and Dangora Morocho. High resolution Sentinel-2 and very high resolution PlanetScope satellite images are used. The development of the classification algorithm is done in the Google Earth Engine cloud based environment. Ten combinations of the 4 most important parameters of the Random Forest classification method are tested. The defined legend contains 8 land cover classes, namely built-up, crops, grassland/pasture, forest, scrubland, bareland, wetland and water body. The training dataset is collected in the field during the fall 2020. The classification results of the two data types at two scales are compared. The highest overall accuracy for land cover classification of Sidama region came out to be 84.1% and kappa index of 0.797, with Random Forest method parameters of 100 trees, 4 spectral bands entering each tree, value of 1 for leaf population and 40% of training data used for each tree. For the land cover classification of Chancho and Dangora Morocho kebele with the same method settings, the overall accuracy came out to be 66.00 and 73.73% and kappa index of 0.545 and 0.601. For the classification of Chancho kebele, a different combination of parameters (80, 3, 1, 0.4) worked out better...
|
|
Data Analysis of a Company Producing Medical Supplies
Kulhánková, Monika ; Bartík, Vladimír (referee) ; Burgetová, Ivana (advisor)
This bachelor's thesis deals with the analysis of the company's sales data, specifically the classification of the customer's type according to his sales data. It provides a theoretical introduction to data mining. It describes the classification process and methods for creating classifiers and presents the CRISP-DM model. This thesis describes the provided data sets, from which the relevant attributes are selected. The data are preprocessed and used in the creation and testing of classification models. The result of this thesis is a comparison of the achieved results.
|
|
Sentiment Analysis of Czech and Slovak Social Networks and Web Discussions
Sojka, Matěj ; Dočekal, Martin (referee) ; Smrž, Pavel (advisor)
Thanks to digitalization, the spread of opinions in the population has accelerated sharply in the recent years, however the need to understand them has not changed. The goal of this thesis was to create a system for automatic data collection from social media and web discussions and sentiment analysis in Czech and Slovak language. The system has a web interface for visualizing results and configuring data analysis. The system is capable of offering topics to the user that it considers to occur in the selected data and group posts based on user-defined opinions.
|