Suchergebnisse

This master work describes the most widely used artificial intelligence methods and the possibilities to apply them in credit risk evaluation which is one of the most important fields in banking and in finance. The main problem here is to evaluate the risk arising when a creditor gives a credit to a particular individual or an enterprise, using various mathematical, statistical or other methods and techniques. This risk arises when the debtor isn't able to pay for the loan to the creditor in time which means additional loss. It can appear in many forms depending on the type of debtor (individ-ual, enterprise, government of an abroad country) and type of financial instrument or action that is done with it (giving of a loan, transactions of financial derivatives, etc.), this is the reason why fi-nancial institutions and for it's evaluation and management use various different methodologies which comprise a lot of methods and techniques from credit scoring (evaluating by a particular formula, usually linear) and evaluating different factors, like management and business strategies or policies, to classification by various criterions by using modern and sophisticated methods, either algebraic, either artificial intelligence and machine learning. This field is widely researched and many new techniques are being found. The research here is concentrated mainly on Support Vector Machines (abbr. SVM) which is one of the most popular artificial intelligence and machine learning techniques and whose effectiveness has been widely proved. This research is done in order to investigate the possibilities to adapt SVM method to the problem described above, and to implement a system which uses one of SVM techniques.

This master work describes the most widely used artificial intelligence methods and the possibilities to apply them in credit risk evaluation which is one of the most important fields in banking and in finance. The main problem here is to evaluate the risk arising when a creditor gives a credit to a particular individual or an enterprise, using various mathematical, statistical or other methods and techniques. This risk arises when the debtor isn't able to pay for the loan to the creditor in time which means additional loss. It can appear in many forms depending on the type of debtor (individ-ual, enterprise, government of an abroad country) and type of financial instrument or action that is done with it (giving of a loan, transactions of financial derivatives, etc.), this is the reason why fi-nancial institutions and for it's evaluation and management use various different methodologies which comprise a lot of methods and techniques from credit scoring (evaluating by a particular formula, usually linear) and evaluating different factors, like management and business strategies or policies, to classification by various criterions by using modern and sophisticated methods, either algebraic, either artificial intelligence and machine learning. This field is widely researched and many new techniques are being found. The research here is concentrated mainly on Support Vector Machines (abbr. SVM) which is one of the most popular artificial intelligence and machine learning techniques and whose effectiveness has been widely proved. This research is done in order to investigate the possibilities to adapt SVM method to the problem described above, and to implement a system which uses one of SVM techniques.

In our paper we present a corpus of transcribed Lithuanian parliamentary speeches. The corpus is prepared in a specific format, appropriate for different authorship identification tasks. The corpus consists of approximately 111 thousand texts (24 million words). Each text matches one parliamentary speech produced during an ordinary session from the period of 7 parliamentary terms starting on March 10, 1990 and ending on December 23, 2013. The texts are grouped into 147 categories corresponding to individual authors, therefore they can be used for authorship attribution tasks; besides, these texts are also grouped according to age, gender and political views, therefore they are also suitable for author profiling tasks. Whereas short texts complicate recognition of author speaking style and are ambiguous in relation to the style of other authors, we incorporated only texts containing not less than 100 words into the corpus. In order to make each category as comprehensive and representative as possible, we included only those authors, who produced speeches at least 200 times. All the texts are lemmatized, morphologically and syntactically annotated, tokenized into the character n-grams. The statistical information of the corpus is also available. We have also demonstrated that the created corpus can be effectively used in authorship attribution and author profiling tasks with supervised machine learning methods. The corpus structure also allows using it with unsupervised machine learning methods and can be used for creation of rule-based methods, as well as in different linguistic analyses.

In our paper we present a corpus of transcribed Lithuanian parliamentary speeches. The corpus is prepared in a specific format, appropriate for different authorship identification tasks. The corpus consists of approximately 111 thousand texts (24 million words). Each text matches one parliamentary speech produced during an ordinary session from the period of 7 parliamentary terms starting on March 10, 1990 and ending on December 23, 2013. The texts are grouped into 147 categories corresponding to individual authors, therefore they can be used for authorship attribution tasks; besides, these texts are also grouped according to age, gender and political views, therefore they are also suitable for author profiling tasks. Whereas short texts complicate recognition of author speaking style and are ambiguous in relation to the style of other authors, we incorporated only texts containing not less than 100 words into the corpus. In order to make each category as comprehensive and representative as possible, we included only those authors, who produced speeches at least 200 times. All the texts are lemmatized, morphologically and syntactically annotated, tokenized into the character n-grams. The statistical information of the corpus is also available. We have also demonstrated that the created corpus can be effectively used in authorship attribution and author profiling tasks with supervised machine learning methods. The corpus structure also allows using it with unsupervised machine learning methods and can be used for creation of rule-based methods, as well as in different linguistic analyses.

In our paper we present a corpus of transcribed Lithuanian parliamentary speeches. The corpus is prepared in a specific format, appropriate for different authorship identification tasks. The corpus consists of approximately 111 thousand texts (24 million words). Each text matches one parliamentary speech produced during an ordinary session from the period of 7 parliamentary terms starting on March 10, 1990 and ending on December 23, 2013. The texts are grouped into 147 categories corresponding to individual authors, therefore they can be used for authorship attribution tasks; besides, these texts are also grouped according to age, gender and political views, therefore they are also suitable for author profiling tasks. Whereas short texts complicate recognition of author speaking style and are ambiguous in relation to the style of other authors, we incorporated only texts containing not less than 100 words into the corpus. In order to make each category as comprehensive and representative as possible, we included only those authors, who produced speeches at least 200 times. All the texts are lemmatized, morphologically and syntactically annotated, tokenized into the character n-grams. The statistical information of the corpus is also available. We have also demonstrated that the created corpus can be effectively used in authorship attribution and author profiling tasks with supervised machine learning methods. The corpus structure also allows using it with unsupervised machine learning methods and can be used for creation of rule-based methods, as well as in different linguistic analyses.

In our paper we present a corpus of transcribed Lithuanian parliamentary speeches. The corpus is prepared in a specific format, appropriate for different authorship identification tasks. The corpus consists of approximately 111 thousand texts (24 million words). Each text matches one parliamentary speech produced during an ordinary session from the period of 7 parliamentary terms starting on March 10, 1990 and ending on December 23, 2013. The texts are grouped into 147 categories corresponding to individual authors, therefore they can be used for authorship attribution tasks; besides, these texts are also grouped according to age, gender and political views, therefore they are also suitable for author profiling tasks. Whereas short texts complicate recognition of author speaking style and are ambiguous in relation to the style of other authors, we incorporated only texts containing not less than 100 words into the corpus. In order to make each category as comprehensive and representative as possible, we included only those authors, who produced speeches at least 200 times. All the texts are lemmatized, morphologically and syntactically annotated, tokenized into the character n-grams. The statistical information of the corpus is also available. We have also demonstrated that the created corpus can be effectively used in authorship attribution and author profiling tasks with supervised machine learning methods. The corpus structure also allows using it with unsupervised machine learning methods and can be used for creation of rule-based methods, as well as in different linguistic analyses.

In our paper we present a corpus of transcribed Lithuanian parliamentary speeches. The corpus is prepared in a specific format, appropriate for different authorship identification tasks. The corpus consists of approximately 111 thousand texts (24 million words). Each text matches one parliamentary speech produced during an ordinary session from the period of 7 parliamentary terms starting on March 10, 1990 and ending on December 23, 2013. The texts are grouped into 147 categories corresponding to individual authors, therefore they can be used for authorship attribution tasks; besides, these texts are also grouped according to age, gender and political views, therefore they are also suitable for author profiling tasks. Whereas short texts complicate recognition of author speaking style and are ambiguous in relation to the style of other authors, we incorporated only texts containing not less than 100 words into the corpus. In order to make each category as comprehensive and representative as possible, we included only those authors, who produced speeches at least 200 times. All the texts are lemmatized, morphologically and syntactically annotated, tokenized into the character n-grams. The statistical information of the corpus is also available. We have also demonstrated that the created corpus can be effectively used in authorship attribution and author profiling tasks with supervised machine learning methods. The corpus structure also allows using it with unsupervised machine learning methods and can be used for creation of rule-based methods, as well as in different linguistic analyses.

Cryptocurrency prices have high volatility compared to traditional financial time series. This is the reason why it is hard to predict prices properly using standard models. The aim of this work is to apply the hidden Markov chain model to predict cryptocurrency price direction and to determine regime switching impact to cryptocurrency prices. Various political actions and economic regimes affect cryptocurrency prices. Logarithmic returns used for 2-, 3- ,4- ,5-state hidden Markov model application. It was found that 3-state model is the best to find hidden regimes in cryptocurrency price changes. This model divided cryptocurrency returns according to their volatility into low, medium, and extreme volatility regimes. For the price direction forecast the hidden Markov model was constructed by discretizing the cryptocurrency prices returns into 5 intervals. Based on them, a sequence of observations was formed for 3-year data. The assumption was made that there are two hidden states which represents increasing and decreasing prices. Baum-Welch algorithm was used for learning the model. Then predicted price directions for 3 months data and calculated the accuracy of the model, which was compared with machine learning methods (random forest algorithm, artificial neural network method, and naive Bayesian classifier). In average the prediction made by the hidden Markov model is more accurate than other methods when daily observations are considered.

Cryptocurrency prices have high volatility compared to traditional financial time series. This is the reason why it is hard to predict prices properly using standard models. The aim of this work is to apply the hidden Markov chain model to predict cryptocurrency price direction and to determine regime switching impact to cryptocurrency prices. Various political actions and economic regimes affect cryptocurrency prices. Logarithmic returns used for 2-, 3- ,4- ,5-state hidden Markov model application. It was found that 3-state model is the best to find hidden regimes in cryptocurrency price changes. This model divided cryptocurrency returns according to their volatility into low, medium, and extreme volatility regimes. For the price direction forecast the hidden Markov model was constructed by discretizing the cryptocurrency prices returns into 5 intervals. Based on them, a sequence of observations was formed for 3-year data. The assumption was made that there are two hidden states which represents increasing and decreasing prices. Baum-Welch algorithm was used for learning the model. Then predicted price directions for 3 months data and calculated the accuracy of the model, which was compared with machine learning methods (random forest algorithm, artificial neural network method, and naive Bayesian classifier). In average the prediction made by the hidden Markov model is more accurate than other methods when daily observations are considered.

Cryptocurrency prices have high volatility compared to traditional financial time series. This is the reason why it is hard to predict prices properly using standard models. The aim of this work is to apply the hidden Markov chain model to predict cryptocurrency price direction and to determine regime switching impact to cryptocurrency prices. Various political actions and economic regimes affect cryptocurrency prices. Logarithmic returns used for 2-, 3- ,4- ,5-state hidden Markov model application. It was found that 3-state model is the best to find hidden regimes in cryptocurrency price changes. This model divided cryptocurrency returns according to their volatility into low, medium, and extreme volatility regimes. For the price direction forecast the hidden Markov model was constructed by discretizing the cryptocurrency prices returns into 5 intervals. Based on them, a sequence of observations was formed for 3-year data. The assumption was made that there are two hidden states which represents increasing and decreasing prices. Baum-Welch algorithm was used for learning the model. Then predicted price directions for 3 months data and calculated the accuracy of the model, which was compared with machine learning methods (random forest algorithm, artificial neural network method, and naive Bayesian classifier). In average the prediction made by the hidden Markov model is more accurate than other methods when daily observations are considered.

Cryptocurrency prices have high volatility compared to traditional financial time series. This is the reason why it is hard to predict prices properly using standard models. The aim of this work is to apply the hidden Markov chain model to predict cryptocurrency price direction and to determine regime switching impact to cryptocurrency prices. Various political actions and economic regimes affect cryptocurrency prices. Logarithmic returns used for 2-, 3- ,4- ,5-state hidden Markov model application. It was found that 3-state model is the best to find hidden regimes in cryptocurrency price changes. This model divided cryptocurrency returns according to their volatility into low, medium, and extreme volatility regimes. For the price direction forecast the hidden Markov model was constructed by discretizing the cryptocurrency prices returns into 5 intervals. Based on them, a sequence of observations was formed for 3-year data. The assumption was made that there are two hidden states which represents increasing and decreasing prices. Baum-Welch algorithm was used for learning the model. Then predicted price directions for 3 months data and calculated the accuracy of the model, which was compared with machine learning methods (random forest algorithm, artificial neural network method, and naive Bayesian classifier). In average the prediction made by the hidden Markov model is more accurate than other methods when daily observations are considered.

The Fourth Industrial Revolution and the orientation towards climate change neutrality have fundamentally changed the way we look at competitiveness. Global challenges have increased interest in countries' competitiveness. Thus, in the final master's project the competitiveness of the European Union countries is evaluated according to the most important determinants, ecological footprint and the countries are clustered according to the obtained analysis results. In the literature review the latest concept of competitiveness is presented, 9 groups of competitiveness factors and 52 factors-indicators are identified. The missing values are filled in by MICE method, and after evaluating the correlation, 42 factors-indicators are further used in the study. Three methods are used for clustering - hierarchical Ward linkage method, K-means and machine learning method t-SNE. After the cluster analysis, the countries are clustered into four groups. Neighboring countries have been found to be most similar in terms of competitiveness. Country cluster profiles are interpreted using the results of the PCA method. According to the identified groups of factors, the competitiveness index developed by the author of the project revealed that the most competitive EU countries are Sweden, Germany and the Netherlands. The least competitive countries are Romania, Bulgaria and Greece. Assessing the link between the ecological footprint and competitiveness, positive progress is seen in the EU. In addition, EU countries are moving from low competitiveness and low ecological footprint to high competitiveness and low ecological footprint.