Abstract
dc:descriptionIn recent decades, there has been a noticeable increase in the presence of event sequences as a form of data. This surge is mainly due to the rise and widespread use of Internet of Things (IoT) devices, which continuously collect data by observing their surroundings and detecting changes. These large volumes of event sequences are stored in databases to facilitate various forms of analysis, enabling organizations to uncover valuable insights. This thesis addresses several critical challenges in efficiently managing and analyzing such data. More specifically, this thesis introduces SIESTA, an infrastructure designed to facilitate efficient pattern analysis over large event logs by leveraging inverted indexing techniques. Pattern analysis includes the detection of arbitrary patterns and the exploration of potential continuations for unfinished event sequences. SIESTA comprises two separate components: (a) the preprocessing component that incrementally builds the indices utilizing the massive parallelization offered by the Apache Spark framework, storing them in Cassandra, a scalable key-value database; and (b) the query processor, which uses these indices to prune the search space and expedite response times of the pattern queries. Various alternative approaches in both components have been investigated to optimize performance. Extensive evaluation demonstrates that SIESTA achieves significantly lower response times compared to state-of-the-art systems, confirming its effectiveness and scalability in big data scenarios. SIESTA's capabilities are further extended to establish it as a comprehensive, scalable, and trustworthy tool for cloud-native pattern detection with enhanced expressiveness. By integrating a Complex Event Processing (CEP) engine, specifically SASE, within SIESTA's query processor, the system can efficiently handle more complex patterns, including those with negations and iterations, while also providing explanations for unexpected result sets. Additionally, alternatives for storage solutions offering a more cost-efficient solution for cloud deployment are explored. The evaluation of these enhancements highlights SIESTA's superior scalability and pattern detection capabilities compared to competitors like Elasticsearch, FlinkCEP, and MATCH_RECOGNIZE. The effective pruning mechanisms and inherent parallelization of SIESTA enabled its adaptation to mine declarative constraints in big data scenarios. Specifically, the query processor was extended while the index building remained unchanged, making this an extension to SIESTA's functionalities without losing any of the previously supported queries. Through thorough evaluation using real-world datasets, the proposed solution demonstrates superior performance compared to existing ad-hoc solutions from the Business Process Management (BPM) field, such as Declare Miner and MINERful, particularly when multiple support thresholds are considered. Finally, this thesis investigates the detection of temporal anomalies in business process event logs. Four distinct proximity-based techniques tailored to detect temporal anomalies at both the event and trace levels are introduced. Additionally, the impact of different distance metrics on these techniques is investigated. The conclusions drawn from extensive experimentation offer guidance on selecting the appropriate proximity technique based on the type of anomalies being targeted and the characteristics of the dataset.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Mavroudopoulos, Ioannis
- Μαυρουδόπουλος, Ιωάννης
Subjects
dc:subject × 16- Ακολουθίες συμβάντων
- Ανάλυση μοτίβων
- Μεγάλα δεδομένα (BigData)
- Κλιμακωσιμότητα
- Event sequences
- Pattern analysis
- Big data
- Scalability
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Επιστήμη ηλεκτρονικών υπολογιστών και Πληροφορική, άλλοι τομείς
- Πληροφοριακά συστήματα
- Computer and Information Sciences
- Natural Sciences
- Computer and Information sciences, miscellaneous
- Information Systems
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/57176