University of Patras
Σχεδίαση και υλοποίηση προηγμένων αλγορίθμων μηχανικής μάθησης για την ανάλυση και πρόγνωση χρονοσειρών
Abstract
dc:descriptionThis dissertation focuses on the application and development of advanced Machine Learning algorithms for the analysis and prediction of time series, with applications in various sectors such as smart cities, public health, and financial technology. The main objectives of the dissertation are the application of existing Machine Learning techniques to new time series datasets and the development of innovative algorithms to improve the analysis and prediction of these data. It begins with a detailed overview of Deep Learning techniques in multivariate time series, exploring applications for smart cities. This study fills a gap in the literature by providing a clear picture of the opportunities and challenges encountered in applying these methods. The analysis and prediction of the COVID-19 pandemic through multivariate time series is the next focus. Through the comparison of Traditional Statistical methods and advanced Machine Learning models, the dissertation offers a deeper understanding of the capabilities of these algorithms in managing health data. In the field of financial technology, the dissertation examines Bitcoin price prediction by incorporating sentiment analysis from tweets, showcasing the capabilities of Deep Learning in analyzing complex and new data sources. Additionally, the dissertation focuses on the application of the Cascade technique for processing multivariate time series data. This approach proves particularly effective in improving prediction accuracy, combining various Machine Learning methods to achieve the best possible performance. Finally, the development of a Hybrid algorithm that combines the statistical ARIMA method with the advanced Deep Learning algorithm Temporal Fusion Transformer demonstrates the value of innovation in the analysis of environmental data and enhances the applicability of Machine Learning techniques. This dissertation presents a comprehensive effort to improve the capability of Machine Learning to handle complex and diverse types of time series, leveraging existing techniques and developing innovative solutions. By incorporating advanced analysis methods and creating new approaches, this work attempts to bring significant improvements in the prediction and interpretation of time series, offering additional real-life applications and helping to address practical challenges in numerous scientific and technological fields.
Degree
thesis:*- Grantor dc:publisher
- University of Patras
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Linardatos, Pantelis
- Λιναρδάτος, Παντελής
Subjects
dc:subject × 10Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/57228