Abstract
dc:descriptionThe ever-increasing amount of environmental information combined with their complexity has grown the need to recruit more sophisticated methods and processes to effectively approach them. The newly emerging field of Environmental Data Science, which is a combination of environmental science, machine learning, data mining, and informatics, needs now, more than ever, our attention, as our planet is facing serious challenges.This Dissertation gives an overview of how the recent advances in data science and machine learning can be successfully applied to solve environmental problems. It tries to combat the drawbacks of the current literature in the domain, where there are no standard and efficient pipelines, data are limited, reproducibility is low, and convenient machine learning algorithms are used. Efficient pipelines are constructed in order to collect, merge and even create datasets, pre-process, and analyze them. Supervised and unsupervised algorithms are employed in order to model, validate, visualize and interpret a plethora of environmental use cases.A significant focus was given to understanding and explaining the machine learning black-box through feature selection and exploratory data analysis. This explainability of the hidden properties of the data is essential in this domain, since it is dealing with real-world phenomena that affect the quality of human lives. Feature selection methods that perform much higher than the baseline are found to be working on environmental data including both individual and ensemble methods. Using these methods, critical insights, related to both data science and environment, are presented regarding seagrass species detection and distribution modeling of fish species in the Mediterranean Sea.Climate change is investigated from the perspective of textual micro-blogging data, where the human opinions on the topic are studied from multiple dimensions. A novel climate change dataset is created using state-of-the-art machine and deep learning models, and is made publicly available. Using this comprehensive dataset in space, time and volume, it was possible to extract valuable information using geolocation, human gender, stance, sentiment, historic temperatures, disasters, and topics.
Degree
thesis:*- Grantor dc:publisher
- Democritus University of Thrace (DUTH)
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Effrosynidis, Dimitrios
- Ευφροσυνίδης, Δημήτριος
Subjects
dc:subject × 10- Περιβαλλοντική Επιστήμη Δεδομένων
- Μηχανική μάθηση
- Environmental Data Science
- Machine learning
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Επιστήμες Μηχανικού και Τεχνολογία
- Υπολογιστές, Υλικό (hardware) και Αρχιτεκτονική
- Electrical Engineering, Electronic Engineering, Information Engineering
- Engineering and Technology
- Computer science, Hardware and Architecture
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/52631