Friday 07 March 2025
A new approach to identifying anomalies in time series data has been proposed, one that focuses on synchronizing anomalies across multiple variables. The method, called Synchronised Anomaly Agreement Index (SAAI), uses a novel combination of dynamic time warping and clustering algorithms to identify patterns in multivariate time series data.
The problem of anomaly detection is a challenging one, particularly when dealing with complex systems like those found in industrial control systems or financial markets. Traditional methods often rely on machine learning algorithms that are trained on labeled data, but these can be prone to overfitting and may not generalize well to new, unseen data.
SAAI takes a different approach by focusing on the relationships between anomalies across multiple variables. By aligning anomalies in time and identifying patterns of synchronization, SAAI is able to identify clusters of similar anomalies that may have been missed by traditional methods.
The algorithm works by first identifying anomalous subsequences in each variable using dynamic time warping. These subsequences are then compared across variables to identify synchronized anomalies. The quality of the clustering solution is evaluated using a novel metric, called SAAI, which takes into account both the similarity between clusters and the number of pseudo-clusters (i.e., clusters with only one element).
In experiments on synthetic data, SAAI outperformed traditional methods like k-means and silhouette score in identifying the true number of anomaly classes. The algorithm also performed well on real-world data from an industrial control system, correctly identifying 10 clusters of anomalies.
One of the key advantages of SAAI is its ability to handle varying lengths of time series data. By using dynamic time warping, the algorithm can align anomalies in different variables even if they have different durations. This makes it particularly well-suited for applications where the length of time series data can vary widely, such as in industrial control systems or financial markets.
The authors of SAAI also explored the impact of varying the weight given to the main term and the penalty term in the SAAI metric. They found that weighting both terms equally (i.e., λ = 0.5) resulted in the best performance, but noted that this may not always be the case depending on the specific application.
In addition to its technical merits, SAAI also has practical applications. For example, it could be used to identify patterns of anomalies in industrial control systems, which could help operators take proactive measures to prevent failures or optimize system performance.
Cite this article: “Synchronizing Anomalies across Multivariate Time Series Data”, The Science Archive, 2025.
Anomaly Detection, Time Series Data, Synchronization, Dynamic Time Warping, Clustering Algorithms, Industrial Control Systems, Financial Markets, Machine Learning, K-Means, Silhouette Score
