Anomaly detection as a one-class problem in discrete event systems / Timo Klerx. Paderborn, 2017
Inhalt
- 1 Introduction
- 2 Fundamentals
- 2.1 Probability Density Functions
- 2.2 System classes
- 2.3 Automata classes
- 2.4 Automata Inference Algorithms
- 2.4.1 Common Elements
- 2.4.2 Learning Untimed Automata
- 2.4.3 Learning Timed Automata
- 2.4.4 Algorithm Analysis Frameworks
- 2.5 One-class Classification
- 3 Related Work
- 3.1 Research Area Overview
- 3.2 Process Mining
- 3.3 Grammatical Inference
- 3.4 Sequence-based Anomaly Detection
- 3.5 Other Anomaly Detection Approaches
- 4 Anomaly Detection with PDTTAs
- 4.1 Motivation
- 4.2 Probabilistic Deterministic Timed Transition Automaton (PDTTA)
- 4.3 Learning PDTTAs
- 4.3.1 PDTTA Learning Algorithm ProDTTAL
- 4.3.2 Runtime Complexity
- 4.3.3 Convergence of ProDTTAL
- 4.3.4 Properties of Timed Automata Inference Algorithms
- 4.4 Anomaly Detection
- 4.5 Anomaly Detection in a Two-/Multi-class Setting
- 5 Anomaly Detection Evaluation
- 5.1 Performance Metrics
- 5.2 The Curse of One-class Evaluation
- 5.2.1 Anomalies in Discrete Event Systems
- 5.2.2 Random anomalies
- 5.2.3 Model-based simulated anomalies
- 5.2.4 Anomaly Rate
- 5.3 Experiment Design / Scaling of Experiments
- 6 Experimental Evaluation
- 6.1 Hyperparameter Tuning
- 6.2 Experiment Setup
- 6.3 Preliminary Synthetic Experiments
- 6.4 Synthetic Data Evaluation
- 6.4.1 Direct Anomaly Insertion
- 6.4.2 Additional Results for the Initial PDRTA
- 6.4.3 PDRTA Data Generation
- 6.4.4 PDTTA Data Generation
- 6.4.5 PNTTA Data Generation
- 6.5 Real-world Data Evaluation
- 6.6 Evaluation of Algorithm Scaling
- 7 Conclusion and Future Work
- A Experimental Evaluation
- Acronyms
- Notation
- Bibliography