16168 research outputs found
Sort by
Prediction of cognitive impairment in cardiac surgery using artificial intelligence.
No full textThe master's thesis presents methods for classifying physiological time series data – arterial blood pressure (ABP) and transcranial Doppler (TCD) signals. Signal processing methods (noise removal, segmentation, feature engineering) are reviewed. The most significant ABP and TCD features are identified in predicting postoperative cognitive impairment. During the study, various artificial intelligence models are implemented and compared - from statistical to hybrid and transformer architectures, using spectrogram and recurrence matrices and feature vectors. The accuracy, fit and F1 score of the models are compared. The implemented models and data preparation methods showed that for a small dataset consisting of biological signals of different lengths, the best results were achieved using machine learning models with feature vector input
Forecasting Lithuanian electricity prices on the Nord Pool.
No full textThe final project analyses and forecasts Lithuanian electricity prices for the period from January 1, 2024, to February 28, 2025. This topic is relevant due to the high level of public, business, and political interest in electricity price dynamics, as well as the need to plan activities and costs effectively and to ensure energy efficiency. The following variables have been identified as influencing Lithuanian electricity prices: historical electricity prices in the Swedish SE4 price zone, Poland, and Lithuania; day-ahead electricity consumption forecasts for Lithuania, the Swedish SE4 price zone, Latvia, and Poland; day-ahead electricity generation forecasts for Lithuania, Swedish SE1, SE2, SE3, and SE4 price zones, Finland, Danish DK1 and DK2 price zones, Latvia, Estonia, and Poland; forecasted solar and wind power generation in Lithuania; and commercial inter-system flows between the Swedish SE4 price area and Lithuania, between Latvia and Lithuania, and between Poland and Lithuania, as well as the gas price in Lithuania. Four univariate and four multivariate forecasting models were developed: Random Forest, Support Vector Regression, XGBoost, and LightGBM. The best-performing univariate forecasting model is the Random Forest, with parameters ntree = 300, mtry = 2, and nodesize = 5, achieving an average MAPE of 0.05%. The best-performing multivariate prediction model is XGBoost, with parameters eta = 0.05, max_depth = 6, subsample = 0.8, and colsample_bytree = 0.8. However, the multivariate prediction model did not outperform the univariate one — the XGBoost model achieved an average MAPE of 28.12%
Research on the impact of financial statement releases on stock price volatility.
No full textThis master's thesis examines the impact of corporate financial disclosures on stock return volatility, aiming to determine whether investor reactions depend on the nature of the information, the timing of disclosure, and the broader economic context. The study is grounded in information asymmetry and behavioral finance theories, which suggest that market participants respond unevenly to new information and often make decisions that deviate from full rationality. The empirical analysis applies the Event Study methodology to a sample of 153 S&P 500 companies over the period 2014–2024. The study evaluates investor responses to both quarterly and annual earnings announcements, with groupings based on industry sector, economic cycle phase, day of the week, and earnings surprise (EPS) classification—categorized as positive, negative, or neutral. Several event windows were used, ranging from short-term [–1;+1], [–2;+2] to longer horizons [+1;+10],[+1;+30], and [+1;+60] to capture both immediate and delayed market responses. The analysis includes the calculation of abnormal returns (AR), cumulative abnormal returns (CAR), and cumulative average abnormal returns (CAAR). Statistical testing employed non-parametric methods (Mann–Whitney, Wilcoxon, Fligner–Killeen), logistic and quantile regressions, as well as entropy-based measures to assess the variation in reaction patterns. The results show that investors respond most strongly to negative earnings news. Sector affiliation and the economic cycle influence the nature of these reactions, and earnings announcements released on Fridays trigger significantly more negative short-term responses. Quarterly reports more frequently lead to stronger and more dispersed stock reactions compared to annual reports. The findings confirm that investor behavior is asymmetric and context-dependent, offering new insights into how financial information is incorporated into stock prices. The study contributes to both academic research and practical applications in market analysis
Interactive learning tools to improve the education of pre-school children.
No full textThe final Master's project explores the possibility of using interactive learning tools in the educational process of pre-school children in order to improve the quality of learning, to stimulate children's activity, engagement and motivation to learn. Contemporary trends in education highlight the need to use innovative, child-friendly and attractive tools that are integrated into the content of education and that correspond to children's developmental characteristics. In this context, the aim of this work was not only to theoretically justify the importance of interactive tools, but also to develop and practically test the virtual environment "priesmokyklinis.com", which is adapted to pre-school education. The object of the study is interactive tools and their application in pre-school education. The aim of the study is to improve the quality of children's education and their learning outcomes through the use of interactive tools integrated in a virtual environment. The methods of the research are: analysis of scientific literature, written survey, statistical data analysis. The project developed a virtual environment "priesmokyklinis.com" with integrated interactive teaching tools - original presentations, tasks, educational videos, printable tools and games, adapted for the education of pre-school children, both contact and distance. The study involved pre-school teachers in Jurbarkas district, who tested the virtual environment and provided feedback on its use in practice. According to the data, the environment is assessed as clear, user-friendly, functional and in line with the curriculum requirements. The majority of respondents (59%) indicated that the use of the environment significantly increased children's engagement in the activities, and 53% of the teachers confirmed that the virtual activities helped children to better understand the content. The virtual environment was particularly useful in developing numeracy, language, concentration and social skills. 100% of respondents indicated that they would continue to use the system and would recommend it to colleagues. The survey also identified areas for improvement: a greater variety of tasks according to children's abilities, more video lessons, audio instructions, the possibility for teachers to create their own tasks, and the development of content for children with special needs. According to the results of the study, the virtual environment is a relevant, useful and purposeful educational tool with practical value in a modern educational context
Žmogaus emocijų, amžiaus ir lyties atpažinimo sistemos InMoov humanoidiniam robotui sukūrimas ir tyrimas.
No full textThis work explores the challenge of real-time facial attribute recognition (emotion, age, gender) for Human-Robot Interaction (HRI), focusing on deployment in resource-limited embedded systems. A full pipeline was developed and tested, including comparisons between deep learning architectures (CNNs, transfer learning models, and a Hybrid CNN-Transformer), optimization with TensorRT, and real-time evaluation on a Jetson Xavier NX using both public datasets and a live experiment with 36 participants. The Hybrid CNN-Transformer offered a good trade-off between performance and efficiency. With TensorRT FP16 optimization, the system reached real-time inference (64 FPS) for all three tasks, achieving up to 26× speedups without sacrificing accuracy. However, the real-time experiment revealed challenges, especially in emotion recognition, where performance dropped due to class imbalance, subtle expressions, and pose variation. Gender and age tasks also showed some biases depending on head pose and demographics. Overall, the thesis shows that real-time multi-task facial analysis on edge devices is feasible, but further work is needed on improving robustness, handling dataset imbalance, and mitigating bias – key steps to enable more perceptive and reliable social robots
Jutiklių matavimais grįsta kietųjų dalelių sklaidos ore analizė intensyvaus eismo miesto teritorijoje.
No full textAir pollution, particularly fine particulate matter 2.5, has a serious impact on human health and urban environmental quality. In response to growing concerns and the need for more spatially detailed air monitoring, this thesis investigates the use of low-cost particulate matter 2.5 sensors to quantify air pollution trends in urban areas of Kaunas city. The study’s goal was to assess the effectiveness of these sensors, compare particulate matter concentrations in various Taikos avenue areas, and discover the link between outdoor and inside pollution levels. Measurements were taken from February to July 2024, with low-cost sensors displayed in six urban areas of Kaunas city. To examine spatial variation and infiltration potential, one sensor was installed outside and one inside at each site. Furthermore, one sensor was co-located with the official Environmental Protection Agency monitoring station to ensure data quality. Descriptive statistical analysis and Pearson correlation test were performed using IBM SPSS Statistics software. The low-cost sensor’s dependability in measuring PM2.5 concentrations has been demonstrated by a good correlation (r = 0,832, p<0,000) when compared to the Environmental Protection Agency’s monitoring station. PM2.5 levels differed significantly among outdoor locations (p<0,001), with the greatest median values reported at monitoring sites near high-traffic routes like Kaunas College and Vytautas Magnus University residential hostel. In terms of indoor and outdoor dynamics, the Kaunas College, first building (r=0,895) and Vytautas Magnus University residential hostel showed the largest correlations, whereas the library (r=0,752) and Kaunas College, second building (r=0,656) had moderate correlations. The medical station (r=0,488) and residential building (0,357) had the weakest indoor and outdoor correlations, which could be attributed to more restricted ventilation and distance from direct traffic pollution. The findings demonstrate that, low-cost sensors are a good tool for dense urban air monitoring and can help to measure spatial differences in air pollution within urban environments. The study demonstrated that proximity to traffic, building orientation, and ventilation characteristics significantly influence PM2.5 infiltration indoors. The approach applied here presents a viable foundation for assessing fine-scale air quality. These results are particularly relevant for cities seeking cost-effective methods to monitor localized exposure and support data-driven air quality management
Evaluation of QT interval estimation methods in wearable electrocardiogram recordings.
No full textEach year, 1 in 100 people dies from sudden cardiac death. One of its indicators is a prolonged QT interval. Cardiologists typically assess the QT interval from electrocardiogram (ECG) signals recorded in clinical settings. To enable early detection of QT prolongation or monitor it when using medications with side effects, ECG monitoring is essential not only in clinics but also beyond them. This creates a need to monitor the QT interval using wearable devices. In this study, we selected three well-known methods for QT interval duration estimation: the tangential, threshold-based, and Philips methods. We evaluated these methods and compared them using signals recorded by the wearable device KTU_Watch. We annotated the QT intervals manually. For additional comparison, we used a reference ECG signals from the MUSIC database. The analysis of signals recorded with the wearable device showed that the tangential method estimates QT interval durations more accurately than the threshold-based or Philips methods. The tangential method achieved a root mean square error (RMSE) of 17.5 ms, with a systematic error of –14.6 ms and a random error of ±19.6 ms. The Philips method reached an RMSE of 22.7 ms, with a systematic error of –5.5 ms and a random error of ±44.5 ms. The threshold-based method was the least accurate, with a 30.2 ms RMSE, a systematic error of 22.7 ms, and a random error of ±40.3 ms. During the study, we observed an issue of incorrect QT interval annotations in the MUSIC reference database, particularly in cases of non-standard ECG morphology caused by inverted T-waves and tachycardia
Introduction: toward futurizing intellectual capital theory and practice /
No full textThis editorial introductory chapter sets the stage for a forward-looking exploration of intellectual capital (IC) in both theoretical and practical contexts. It addresses the evolving landscape of IC in an era marked by rapid technological, institutional and socio-political disruptions and demonstrates the need for futurizing the field of IC. By providing a cohesive framework for the book’s thematic sections, this chapter highlights the novel viewpoints and visions from leading experts concerning future directions for IC research and practice. It also summarizes the book’s 16 main chapters and discusses how each of them pushes the boundaries of conceptual and empirical insights in the field of IC. This comprehensive introduction contributes to deeper understanding of the proposed new paradigms to align IC theory and practice with future economic, social, and technological changes
Collaborative learning, cooperative learning and reflective learning to foster sustainable development: a scoping review /
No full textEducation for Sustainable Development (ESD) plays a key role in addressing global challenges. This study highlights the key role of collaborative, cooperative and reflective learning in enhancing students' ability to promote sustainable development. In this study, narrative analysis techniques were used to analyse 172 manuscripts with the core keywords ‘reflective learning’ or ‘cooperative learning’ or ‘collaborative learning’ and ‘sustainability’ or ‘sustainable development’ and ‘education’ that resulted from the search in SCOPUS and Web of Science for the period 1994 to 2023 and the paper selection process. These articles provide a comprehensive overview of collaborative and reflective learning in the context of sustainable development. This study demonstrates how critical it is to identify and deal with students' limits in terms of reflection and cooperation. To genuinely contribute to sustainable development, students must not only possess the necessary knowledge, but also embody values and global awareness, thus enabling them to reflect on and evaluate their practical experiences within complex contexts. Despite potential obstacles of individual differences and cooperation challenges, the review emphasises the importance of innovative pedagogical strategies in providing students with engaging educational opportunities that inspire a commitment to advocating for sustainable development. This study highlights the potential of diverse pedagogical programmes in cultivating sustainable competencies and underscores the significance of recognising and overcoming constraints for the effective implementation of education in sustainable development
An overview of standard designs developed for public buildings in Lithuania during the 1920s and 1930s /
No full textStandardization of Lithuanian interwar public buildings began in the early 1920s and continued throughout the 1930s. This process was mainly driven by the young state’s need for various types of public buildings, especially in the provinces. To speed up the needed construction, certain types of such buildings began to be constructed according to standard design projects. Thus, the article, through the selected examples, aims to present the development of standardization practices in public architecture which were carried out in Lithuania at that time. It is assumed that despite the ambitions to expand the standardization of public architecture, this was not a smooth process. Thus, while a number of standard designs of public buildings were quite successfully used in practice, there were cases when such designs were not even implemented in practice