# Conformalized density- and distance-based anomaly detection in time-series data

**Authors:** Evgeny Burnaev, Vladislav Ishimtsev  
**arXiv:** [1608.04585](https://arxiv.org/abs/1608.04585)  
**Date:** 2016-08-16  
**Summary page:** https://www.summarizepaper.com/en/paper/1608.04585v1/

## Key points

- Authors Evgeny Burnaev and Vladislav Ishimtsev focus on anomalies in time-series data in critical fields such as healthcare, finance, security, and flight safety.
- They introduce novel algorithms for one-dimensional time-series data that incorporate feature extraction techniques and scoring methods to detect anomalies.
- The algorithms are based on a probabilistic framework rooted in the conformal paradigm.
- By combining density- and distance-based approaches, they provide a comprehensive solution for anomaly detection.
- Feature extraction enhances the algorithm's ability to capture relevant information while the scoring mechanism quantitatively measures anomaly likelihood.
- The probabilistic interpretation adds reliability by assigning confidence levels to detected anomalies.
- Their research contributes valuable insights into anomaly detection methodologies for improving decision-making processes in critical scenarios.

## Summary

In their paper titled "Conformalized Density- and Distance-Based Anomaly Detection in Time-Series Data," authors Evgeny Burnaev and Vladislav Ishimtsev explore the significance of anomalies, or unusual patterns, in time-series data within critical fields such as healthcare, intrusion detection, finance, security, and flight safety. They introduce novel algorithms tailored for one-dimensional time-series data that incorporate feature extraction techniques to identify key characteristics and scoring methods to evaluate the deviation of new observations from existing data points. These algorithms are rooted in a probabilistic framework based on the conformal paradigm for interpreting these scores. By leveraging a combination of density- and distance-based approaches, they offer a comprehensive solution for detecting anomalies in time-series data. The integration of feature extraction enhances the algorithm's ability to capture relevant information from the dataset while the scoring mechanism provides a quantitative measure of anomaly likelihood. The probabilistic interpretation based on conformal prediction adds a layer of reliability to the anomaly detection process by assigning confidence levels to detected anomalies. Overall, Burnaev and Ishimtsev's research contributes valuable insights into anomaly detection methodologies for time-series data analysis. Their innovative approach holds promise for improving decision-making processes in critical scenarios where timely identification of anomalies is crucial for effective risk management and intervention strategies.