Search

The aim of the paper is to construct a universal classifier to discriminate short Lithuanian text excerpts of poetry from that of prose. Here the universality means that the classifier is relatively insensitive to a text content and author's style. Since syllables represent phonetic properties and are less sensitive to text content as compared to words, the classifier training is based on frequencies of syllables in texts to be classified. The text data is taken from digitized library http://ebiblioteka.mkp.emokykla.lt. The error rate of the trained classifier applied to testing excerpts of 100 words is less than 5\%.

Straipsnyje nagrinėjamas atskiras daugelio elementų paskirstymo uždavinio atvejis – studentų dalykų pasirinkimo (paskirstymo) problema. Darbe analizuojama Vilniaus Universiteto bendrųjų universitetinių studijų įgyvendinimo praktika iki 2016 metų. Tiriamosios analizės rezultatai parodo, kad taikomas dalykų paskirstymas pirmumo principu nėra efektyvus. Norint palyginti dalykų paskirstymo algoritmus reikia turėti duomenų aibę su studentų preferencijomis. Darbe, naudojant cenzūruotos regresijos modelį, turimų realių duomenu pagrindu pasiūlytas studentų preferencijų generavimo algoritmas. Imituotų preferencijų duomenys panaudoti lyginant kelių plačiai taikomų praktikoje paskirstymo algoritmų charakteristikas. We study the allocation of courses to students with multi-unit demand. Exploratory analysis of course allocation practices at Vilnius University (VU) is performed. The allocation of students to courses is done by a simple first-come first-served (FCFS) procedure. We show that FCFS procedure does not deliver desirable outcomes. Data with student preferences is needed in order to compare assignment algorithms. We use censored regression model to construct student preferences generation mechanism based on empirical data collected at VU. Using simulated data several multiunit assignment algorithms are compared.

Straipsnio tikslas – įvertinti tekstų statistinį nehomogeniškumą pagal funkcinių žodžių ir kitų lingvistinių elementų vartoseną. Atliktas empirinis tyrimas remiasi mokykloms rekomenduojamų suskaitmenintų grožinės literatūros kūrinių biblioteka http://ebiblioteka.mkp.emokykla.lt. Apskaičiuojami sudarytų dažnų žodžių formų ar kitų kalbinių struktūrų rinkinių, juos vadinsime žymekliais, dažnumai tekstų blokuose, jungiančiuose 50 iš eilės einančių sakinių. Pastebėta, kad žymeklių dažnumai blokuose turi ženklią perteklinę sklaidą, palyginti su lingvistikoje įprastu homogeniškumo modeliu. Pasirinktoms žymeklių grupėms parinkti kelių tipų hierarchiniai binominės logistinės regresijos modeliai, naudojantys autoriaus identifikatorių, bloko ilgį ir likusių žymeklių dažnius blokuose kaip aiškinančiuosius kintamuosius, leido paaiškinti didelę dalį pasirinktųjų žymeklių perteklinės sklaidos. The aim of the paper is to assess the distributional non-homogeneity of texts in the usage of functional words and other linguistic units. Our empirical study is based on recommended school fiction works taken from a digital library at http://ebiblioteka.mkp.emokykla.lt. Sets of frequent word forms, called markers, are made, and their frequency counts in blocks of 50 successive sentences are calculated. The frequency counts of the markers show significant excess variability (overdispersion) with respect to a text homogeneity model usually assumed in linguistics. For chosen markers, different kinds of hierarchical binomial logistic regression models with the author's identifier, the block length and the frequency counts of the remaining markers as explanatory variables are fitted to the block data in order to explain the observed overdispersion of the markers chosen.

Lietuvių kalba pasižymi vaizdingumu, kuris gali būti perteikiamas įvairiomis raiškos priemonėmis. Viena iš jų – metakalbiniai komentarai. Šiame straipsnyje, remiantis VDU sudarytu Lietuvių kalbos tekstynu, statistiškai nagrinėjami vartojamo pasakymo (emocinį) toną apibūdinantys metakalbiniai komentarai rašytinėje lietuvių kalboje. Sudarytas logistinės regresijos modelis, kuriuo siekiama aprašyti, nuo ko priklauso metakalbinio komentaro proporcija tarp visų bazinio žodžio pavartojimų. Lithuanian language is characterized by imagery, which can be conveyed in a variety of expression means. One of them is metalanguage comments. In this article, metalanguage comments that describe the (emotional) tone of saying are analyzed statistically using corpus of contemporary Lithuanian language. A fitted logistic regression model describes effects which influence the proportion of the metalanguage comments among all occurrences of their basic word in a text.

In the paper the problem of adjustment for nonresponse in a WEB survey is addressed. The study is based on Vilnius University student survey on quality assessment of information technology services. A conditional logistic regression model is applied along with traditional nonresponse adjustment methods.

The problem of estimating parameters of Gaussian vector when only its (nonlinear) transformation is observed is addressed. The EM algorithm equations to calculate maximum likelihood estimator are derived. In particular, closed-form formulas of the EM algorithm are derived in the case when only the minimum of two endogenous variables satisfying Gaussian regression model is observed.

Lietuvių kalba yra gana sudėtinga ir lanksti, ir tai gerokai apsunkina efektyvių algoritmų kūrimą automatiniam lietuviškų tekstų apdorojimui. Tekstų stiliaus ypatybių nagrinėjimui buvo pasirinktos universalios kiekybinės charakteristikos, kurios yra nesusijusios su teksto turiniu ir gali būti suskaičiuotos bet kuriam tekstui. Šiame straipsnyje parodoma, kaip matematinės statistikos pagalba galima atskirti ir interpretuoti lietuvių kalbos stilius. Atlikti logtiesinių modelių tyrimai parodo raidinės ir garsinės struktūros sąryšį su moksliniu ir grožiniu stiliumi. Lithuanian language is quite complex and flexible, and its significantly complicates the development of efficient algorithms for the automatic processing of Lithuanian texts. For studying text-styles features were selected the universal quantitative characteristics that are unrelated to the text content and can be calculated for any text. This article shows how mathematical Statistics can help to distinguish and interpret the Lithuanian language styles. Studies of the log-linear models show the connection between the letters and sounds structure and the scientific and fiction.

After an introduction to genetic basics three problems are briefly discussed: microarray data analysis, a definition of noninformative DNA sequence, and genetic sequence alignment. More atention is paid to DNA sequence visualization and regularization of a genome signature.

The main purpose of this paper is to review the significance of the quality in manufacturing and develop a model that forecasts production breakdowns, the research work was carried by using data given by the specific company X, which operates in Lithuania. Machine learning methods such as random forests and binary logistic regression were used writing this paper. Moulding machine specifications and product test results from firm X were utilized in the analysis. The use of random forests provides the following results: parameter27, parameter29, parameter24, parameter25, parameter11, and parameter23 are the key factors influencing the incidence of defects and the prediction accuracy is fairly high. However, the logistic regression model does not perform as planned; it still has to be improved because the projections are inaccurate and unreliable. With more data, company X may enhance these models and put them into practice in the future, saving part of the money set aside for insuring product quality.

At first, in this research paper the definition of labour productivity is reviewed from the theoretical standpoint, together with the academic overview of the development of the information and communication technology (ICT) as well as e-commerce. Further, the academic literature of the relationship between labour productivity and e-commerce is reviewed. Based on such overview, variables which were found in research papers to have an explanatory power on both labour productivity and prominence of e-commerce in a given economy, were chosen. This selected time period of the analysis is from 2008 to 2019, where the quantitative data was taken from the Lithuanian Department of Statistics. The model of the research consists of one dependent and 9 independent variables. The dependent variable is labour productivity which is measured in added value per employed person and is expressed in EUR per hour. The independent variables are the following: 1. Proportion of enterprises which sold (received orders for) goods or services via electronic networks, %; 2. The share of orders received through e-commerce in relation to total turnover, %; 3. Assets, equity and liabilities of enterprises, thousands of Eur; 4. Share of enterprises controlled by legal entities registered abroad, %; 5. Number of enterprises with 1 to 49 employees, units; 6. Number of enterprises with 50 to 249 employees, units; 7. Number of enterprises with more than 250 employees, units; 8. Share of employees using a computer at their workplace, %; 9. Share of employees with access to the internet at their workplace, %. After the correlation analysis, variables that are statistically significant were selected and included in the later panel data models. In order to select the most fit panel data model describing the phenomenon, 3 main models were developed: focused regression, fixed effects, and random effects models. The best model was selected using statistical tests. Using correlation analysis and developing a panel data model, the relationship and its strength between the aforementioned variables were evaluated. The results of the study show that e-commerce and productivity are positively related. Thus, e-commerce has a positive effect on the labour productivity of individual economic sectors and can be considered as one of the factors influencing the increase of labour productivity.

A statistical analysis of the unemployment rate in Lithuania and other economic and social indicators was performed in this bachelor's thesis. The annual 2008-2018 years data of 10 counties of Lithuania is taken from the Lithuanian Department of Statistics database. The R program (R 3.5.0 for Windows) was used to process the data and derive their statistical conclusions. The main goal of bachelor's thesis is to analyse the dependency of unemployment rate from other economic and social indicators. Methods applied for data analysis: correlation analysis, panel linear regression, automatic variable selection methods. The analysis revealed that the changes in unemployment rate in Lithuania is mostly affected by the GDP per capita indicator, gross earnings and the level of higher education.

The thesis is devoted to the linear discriminant analysis of spatially correlated data. The presence of spatial correlation violates the assumption of independent observations which is the background for many classical statistical methods. Therefore, when modelling spatial data it is important to incorporate the spatial correlation, as ignoring it may affect the accuracy of classification procedures. The thesis presents the original discriminant functions based on a plug-in Bayes classification rule, taking into account the spatial correlation between the Gaussian observations. The proposed discriminant functions are applied to assign a single Gaussian random field observation to one of several prescribed populations and to assess the classification risk or error rates. The cases of univariate and multivariate Gaussian random fields are explored. The simulation study confirms that the proposed clasifiers are of a good accuracy and also shows that including spatial correlation gives lower classification risk or error rate.

In this paper we introduce new way of modeling quality of education by estimating university similarity to the top universities. Such quality estimation is based on idea that university study program has tendency to become similar to the top universities study programs. To conduct empirical analysis for tendency testing, economic model is developed. Theoretical model shows that differences in universities' similarities can be explained by differences in professors' salaries. This theoretical insight simplifies empirical work. One only needs cross-sectional data to test the tendency. Empirical analysis showed that tendency is likely to exist. While main idea of theoretical model is that university can achieve approximately same quality of education as top universities by only increasing professors' salaries.

Translated text has certain features which mark it as such, which can be identified using statistical methods. Features such as lexical density, vocabulary richness and word length distribution are some of the marks of translated text identified by existing research. In this work support vector machine models, which were found to be effective for this purpose by previous studies, are applied to corpora of Lithuanian monolingual texts. The models are then augmented using variables constructed to reflect the suggested marks of translated text in an attempt to improve classification performance.

This doctoral dissertation addresses the problem of how to measure a latent process such as globalisation which manifests itself in a large number of statistical indicators. A new globalisation measure is offered. It is based on the evaluation of proportion of future economic growth explained by foreign indicators relative to domestic ones. The construction of this measure is based on the assumption that there is a strong relationship between economic openness or globalisation and economic growth. A method to evaluate the offered globalisation measure is developed so that the design of the model is suited for the structure of the data. Targeted indicators are structured into domestic and foreign blocks and included into a dynamic hierarchical factor model. The model is evaluated employing Markov chain Monte Carlo with Gibbs sampling. The resulting factors are included into economic growth forecast equation and time-varying parameters loading of domestic and foreign factors are evaluated by applying Kalman filter on a dynamic linear model. The application for Lithuanian economy revealed a positive trend in foreign impact on the economy and it depicts the globalisation effect.

This doctoral dissertation addresses the problem of how to measure a latent process such as globalisation which manifests itself in a large number of statistical indicators. A new globalisation measure is offered. It is based on the evaluation of proportion of future economic growth explained by foreign indicators relative to domestic ones. The construction of this measure is based on the assumption that there is a strong relationship between economic openness or globalisation and economic growth. A method to evaluate the offered globalisation measure is developed so that the design of the model is suited for the structure of the data. Targeted indicators are structured into domestic and foreign blocks and included into a dynamic hierarchical factor model. The model is evaluated employing Markov chain Monte Carlo with Gibbs sampling. The resulting factors are included into economic growth forecast equation and time-varying parameters loading of domestic and foreign factors are evaluated by applying Kalman filter on a dynamic linear model. The application for Lithuanian economy revealed a positive trend in foreign impact on the economy and it depicts the globalisation effect.

Forecasting of Financial Indicators The purpose of bachelor thesis was to analyse and solve the problem of delay to submit financial data to the Center of Register. This problem makes it difficult to draw some conclusions about the last year’s financial situation in different sectors. There was used one more source of information - data collected by the Department of Statistics - and created structural model which consists of two regression equations. The idea of obtained model was to forecast medians of financial indicators at any point in time. An efficiency of this method was measured by comparing the forecasts with actual values and was approved in most cases (restrictions of this method are also discussed). It means, that the same idea can be used for predicting values of financial indicators in the future. Thesis consists of 6 parts: introduction, theoretical part, primary data analysis, econometric data analysis, conclusion, references. It has 25 pages without appendix.

This bachelor‘s thesis analysis object are financial time series of airline companies – stock prices returns. In the first part of work is introduced stock price conception, financial time series econometric analysis theory, stylized facts, describing the most commonly applied forecasting models. In the second part is provided stock prices returns econometric analysis, checking whether the investigated data satisfying financial time series stylized facts. Mainly, is checking six most important stylized facts. Also, for each time series selected the most frequently used models for financial data, which gives the best predictions.

In this work we analyse the e-marketing effectiveness of 42 Lithuanian e-shops data. We investigate what factors influence customers choice to buy commodity from newsletter advert while at the same time, the expected number of orders and revenues. In this analysis, we use quantitive and categorical variables related to the information about newsletters and companies. This research based on the negative binomial, zero-inflated and hurdle negative binomial regression models. Searching for the best model, suitable for our data, we use likelihood ratio and Wald criteria for the compariosn of the nested models and Vuong criteria for nonnested models comaprison.

In this master thesis, the famous Thomson Reuters corpus is analyzed using modern statistical models for text data to investigate a novel research question - just how important the content of news is to financial markets? Using topic models, topic mixtures or proportions of topics that news write about within documents (time periods) are inferred and then used to predict market outcomes or test Granger causality between topics and market variables, where topics are treated to be probability distributions of words. Results of the experiments point to the direction that the content of news might not correspond to the market outcomes exactly and do not carry significant amount of predictive power. However, Granger causality testing reveals, that some specific topics, for example regarding Federal Reserve, politics and etc. do seem to Granger cause market variables and especially, the volatility of trading. This seem to suggest that the content of news might be important as a mean for market players to get informed but other factors are at play too which determine how markets are going to behave. Additionally to this research question, inference algorithm using Direct Representation scheme is derived for Supervised Hierarchical Dirichlet Process mixture model which is found to be superior to some other representation schemes in the literature.

The paper presents job satisfaction analysis among Italian entrepreneurs using multivariate statistical techniques. Data are taken from Veneto region (North-East part of Italy) businessmen research started by Department of Statistics of Padova University at 2006. Sample consists of 1216 observations (real population is almost 113 000 entrepreneurs at this region). The outcome variable for the job satisfaction is measured on an ordered, categorical, four-point Likert scale – „dissatisfied‘“, „neither satisfied nor dissatisfied“ , „quite satisfied“ and „satisfied“. Explanatory variables include demographic items, firm characteristics, variables representing the reasons of having started the own business, items associated with work, leisure activities and future perspectives. Quite all variables, as well as the dependent one, beeing categorical, the main objective of this work is to select an appropriate model for such type of data among possible alternatives. Discussions are made on possibility to applicate linear regression with optimal scaling, binary logit for dichotomized dependent variable, multinomial logit for analysis of every single category and ordinal logistic regression with several link functions (logit and cloglog). The number of possible determinants of job satisfaction beeing quite large (there were more than 30 questionnaire items associated with job satisfaction) it was of a great importance choosing which explanatory variables should be included in the models, considering the multicollinearity challenge. In order to reduce the number of variables, it was considered possibility to apply factor analysis. The latter beeing applied only for a quantitative data, an alternative method to extract factors was proposed evaluating polychoric and tetrachoric correlations for ordinal and dichotomious variables instead of Pearson correlations . The results show that considering model assumptions, fit statistics and classification results, the most appropriate model for our data is ordinal logistic regression with cloglog link. Multicollinearity analysis was of a great help to detect some highly associated variables that were transformed or omitted of models. Applying factor analysis the quantity of variables was reduced detecting three factors, that according to the correlated with them variables were named as ‚work‘ ‚leisure‘ and ‚advanced firm‘ factors. Later they all were included in considered models as explanatory variables instead of the variables they represent. Job satisfaction determinants were evaluated by gender. Results show that factors representing leisure and advanced firm both increase job satisfaction for men and women, while working time positively affects women job satisfaction and has negative influence for men job satisfaction. More satisfied with job seem to be unmarried or divorced, high-educated, older men and younger, married women who have not school age children and who have graduated superiour school. It is worth mentioning that analysing reasons of becoming entrepreneur it is seen that ‚professional realization‘ affects possitively job satisfaction for both genders comparing with the reference- ‚starting the autonomous life‘. While ‚money‘ as a reason of haven started the business has positive influence for women and negative one for men comparing with ‚starting autonomous life

Šiame magistro darbe yra pateikiamas įmonės išgyvenamumą prognozuojantis Kokso proporcingųjų rizikų modelis. Pirmoje dalyje yra pateikiama trumpa modelių, susijusių su įmonių bankrotais ar nemokumu, apžvalga. Taip pat yra pristatomi balansinių ataskaitų ir įmonių finansų teoriniai pagrindai, įvairūs finansiniai rodikliai (įmonės pelnas/nuostolis, įsipareigojimai, turtas ir kt.) bei santykiniai finansiniai rodikliai (įmonės bendras skolos rodiklis, grynojo pelningumo koeficientas ir kt.). Be to, pateikiami išgyvenamumo analizės ir Kokso proporcingųjų rizikų modelio teoriniai pagrindai. Antroje darbo dalyje pateikiami empiriniai skaičiavimai. Čia parenkami statistiškai reikšmingi finansiniai ir santykiniai finansiniai rodikliai, įtakojantys įmonės gyvavimo trukmę, sudaromas bei tiriamas Kokso regresijos modelis, kuris vertina įmonės tikimybę išgyventi po tam tikro laiko. Ekonometrinė analizė atliekama naudojant SPSS, MS Excel ir SAS programinę įrangą. This Master thesis develops survival of company prediction Cox proportional hazards model using the statistical methodology of Survival analysis. In the first part the short review of literature about models related with company’s bankruptcy, failure or financial distress is presented. Also the background of financial statement analysis, corporate finance, various financial variables (company’s profit/loss, liabilities, asset and other) and financial ratios (debt ratio, total asset turnover and other) are clarified. Furthermore, the theoretical background of Survival analysis and Cox proportional hazards model are presented. In the following part empirical computations are introduced. There are selected financial variables and financial ratios, which influence company’s survival and are statistically significant. Moreover, Cox regression model for estimating the probability of company’s survival after particular time is presented. Econometric analysis is performed using SPSS, MS Excel and SAS software.

Disertacijos darbo tikslas – įrodyti universalias tiksliąsias nelygybes atsitiktinių dydžių funkcijų nukrypimo nuo vidurkio tikimybėms. Universalios nelygybės pažymi, kad jos yra tolygios pagal tam tikras bendras skirstinių klases ir pagal atsitiktinių dydžių kiekį, kartais ir pagal kitus parametrus. Nelygybės vadinamos tiksliosiomis, jeigu pavyksta sukonstruoti atsitiktinių dydžių seką, kuriai nelygybės virsta lygybėmis. Tokios nelygybės labai naudingos, pavyzdžiui, draudimo matematikoje, konstruojant efektyvius algoritmus. Disertaciją sudaro šeši skyriai. Pirmasis skyrius yra įvadas, kuriame neformaliai pristatomas disertacijoje tiriamas objektas, pateikiamas bendras darbo aprašymas ir motyvacija. Detalesnė kitų autorių rezultatų apžvalga pateikiama atskirai kiekviename skyriuje. Antrasis skyrius skirtas atvejui, kai atsitiktiniai dydžiai yra aprėžti ir simetriniai. Trečiajame skyriuje įrodomos nelygybės atsitiktiniams dydžiams, tenkinantiems dispersijos aprėžtumo sąlygą. Ketvirtajame skyriuje nagrinėjamos sąlyginai aprėžtų atsitiktinių dydžių sumos. Penktajame skyriuje tiriamos atsitiktinių dydžių sekos, sudarančios martingalą arba supermartingalą, ir joms gaunamos universaliosios tikimybinės nelygybės ir sukonstruojama nehomogeninė Markovo grandinė, kuri yra martingalas, ir kuriai minėtos nelygybės virsta lygybėmis. Šeštajame skyriuje rezultatai yra apibendrinami atsitiktinių dydžių sekos Lipšico funkcijoms. The purpose of the dissertation is to prove universal tight bounds for deviation from the mean probability inequalities for functions of random variables. Universal bounds shows that they are uniform with respect to some class of distributions and quantity of variables and other parameters. The bounds are called tight, if we can construct a sequence of random variables, such that the upper bounds are achieved. Such inequalities are useful for example in insurance mathematics, for constructing effective algorithms. We extend the results for Lipschitz functions on general probability metric spaces.

The purpose of the dissertation is to prove universal tight bounds for deviation from the mean probability inequalities for functions of random variables. Universal bounds shows that they are uniform with respect to some class of distributions and quantity of variables and other parameters. The bounds are called tight, if we can construct a sequence of random variables, such that the upper bounds are achieved. Such inequalities are useful for example in insurance mathematics, for constructing effective algorithms. We extend the results for Lipschitz functions on general probability metric spaces. Disertacijos darbo tikslas – įrodyti universalias tiksliąsias nelygybes atsitiktinių dydžių funkcijų nukrypimo nuo vidurkio tikimybėms. Universalios nelygybės pažymi, kad jos yra tolygios pagal tam tikras bendras skirstinių klases ir pagal atsitiktinių dydžių kiekį, kartais ir pagal kitus parametrus. Nelygybės vadinamos tiksliosiomis, jeigu pavyksta sukonstruoti atsitiktinių dydžių seką, kuriai nelygybės virsta lygybėmis. Tokios nelygybės labai naudingos, pavyzdžiui, draudimo matematikoje, konstruojant efektyvius algoritmus. Disertaciją sudaro šeši skyriai. Pirmasis skyrius yra įvadas, kuriame neformaliai pristatomas disertacijoje tiriamas objektas, pateikiamas bendras darbo aprašymas ir motyvacija. Detalesnė kitų autorių rezultatų apžvalga pateikiama atskirai kiekviename skyriuje. Antrasis skyrius skirtas atvejui, kai atsitiktiniai dydžiai yra aprėžti ir simetriniai. Trečiajame skyriuje įrodomos nelygybės atsitiktiniams dydžiams, tenkinantiems dispersijos aprėžtumo sąlygą. Ketvirtajame skyriuje nagrinėjamos sąlyginai aprėžtų atsitiktinių dydžių sumos. Penktajame skyriuje tiriamos atsitiktinių dydžių sekos, sudarančios martingalą arba supermartingalą, ir joms gaunamos universaliosios tikimybinės nelygybės ir sukonstruojama nehomogeninė Markovo grandinė, kuri yra martingalas, ir kuriai minėtos nelygybės virsta lygybėmis. Šeštajame skyriuje rezultatai yra apibendrinami atsitiktinių dydžių sekos Lipšico funkcijoms.