Anomaly detection based on a dynamic Markov model.

Area of Science:

• Computer Science
• Data Science
• Cyber Security

Background:

• Anomaly detection in sequence data is crucial for applications like fraud and intrusion detection.
• Traditional Markov models have limitations, including short memory or reduced reliability in higher-order models.
• Existing models struggle to capture sequences with evolving trends.

Purpose of the Study:

• To propose a novel dynamic Markov model for robust anomaly detection in sequence data.
• To address the limitations of classical and higher-order Markov models in handling data trends and interactions.
• To enhance the adaptability and stability of anomaly detection systems.

Main Methods:

• Sequence data is segmented using a sliding window approach.
• A higher-order Markov model is established within each window, defining states based on data values.
• An anomaly substitution strategy is implemented to maintain model integrity and continuous detection.

Main Results:

• The dynamic Markov model effectively balances memory length and trend tracking in sequence data.
• The anomaly substitution strategy prevents detected anomalies from compromising model performance.
• Experimental results show improved adaptability and stability in anomaly detection using simulated and real-world datasets.

Conclusions:

• The proposed dynamic Markov model offers a significant improvement over traditional methods for anomaly detection.
• This approach enhances the reliability and accuracy of anomaly detection in diverse sequence data applications.
• The method provides a stable and adaptable solution for continuous anomaly detection in dynamic environments.