Dimensionality reduction for complex event forecasting
Μείωση διαστάσεων για πρόγνωση σύνθετων γεγονότων
View/
Keywords
CEF ; Complex event forecasting ; Artificial Inteligence ; Neural networks ; Dimensionality reduction ; Feature selection ; Feature importance ; AIAbstract
Complex Event Recognition and Forecasting (CER/F) has emerged as a pivotal area in artificial intelligence, addressing the need to detect, understand, and predict intricate patterns in dynamic and high-volume data streams. This thesis explores a neuro-symbolic approach to CER/F by integrating dimensionality reduction methods with the Wayeb framework—an automata-based system designed for efficient forecasting. The study focuses on dimensionality reduction as a key strategy to enhance model performance and interpretability, particularly through feature selection methods that identify and retain the most relevant attributes. By reducing the dimensionality of the input data, the alphabet fed into the automaton is simplified, enabling more efficient computations and improved accuracy in CER/F tasks, while preserving the interpretive integrity of symbolic reasoning. To validate the approach, extensive experiments are conducted using synthetic datasets. The impact of dimensionality reduction on recognition and forecasting accuracy, runtime efficiency, and model interpretability is thoroughly evaluated. Results indicate that feature selection significantly improves the scalability of the Wayeb framework and facilitates better generalization in forecasting complex events.
Postgraduate Studies Programme
Τεχνητή Νοημοσύνη - Artificial IntelligenceDepartment
Σχολή Τεχνολογιών Πληροφορικής και Επικοινωνιών. Τμήμα Ψηφιακών ΣυστημάτωνCorporate Department
National Center of Scientific Research "Demokritos"Number of pages
87Language
EnglishCollections
Except where otherwise noted, this item's license is described as
Αναφορά Δημιουργού-Μη Εμπορική Χρήση 3.0 Ελλάδα
Αναφορά Δημιουργού-Μη Εμπορική Χρήση 3.0 Ελλάδα