Institutional Repository of Vilnius University
Anomalijų paieška finansinių duomenų eilutėse /
Abstract
dc:descriptionAnomaly detection and its algorithms are widely used in Medicine, Fraud detection, Bioinformatics, and other fields. However, anomaly detection in financial time series, especially stocks, has not yet been widely studied. This thesis aims to explore the algorithms used for anomaly detection and use other researches to find out why anomaly detection algorithms are rarely used. Moreover, find an algorithm suitable for detecting anomalies in stock prices and find the best anomaly detection algorithm parameters for detecting anomalies in stock prices. To achieve this goal, a new evaluation technique of anomaly detection models was developed using news headlines about stocks. The experiment used 10 stocks that had the biggest correlation coefficient between absolute change in stock price and news count about the stock. Days, when the algorithm marked as anomalous and the news count was unusually high, were counted, and using this count anomaly detection algorithms were compared and evaluated. Using the proposed evaluation technique, on average, the best result with default algorithm parameters was produced by the KNN anomaly detection algorithm. The best result with optimal parameters was produced by the OCSVM algorithm with kernel parameter values linear and poly. However, the experiment led to the conclusion that there is no one algorithm that is most suitable for all stocks.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Nostytė, Vaiva,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD146237024&prefLang=en_US