Time-Series Seasonality Detection Through Subsequence Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems struggle with detecting anomalies in time-series data due to the complexity and computational intensity of identifying seasonality patterns, leading to overwhelming alerts, false positives, and false negatives, especially when dealing with vast amounts of raw data.
Innovation Solution
A system and methodology that includes analyzing time-series data for regularity, performing data aggregation based on data points' intervals, determining seasonality patterns through clustering, and generating anomaly bands using silhouette scores and thresholds to identify anomalies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If seasonality patterns are detected using traditional methods, then anomaly detection capability is improved, but computational complexity and processing time increase significantly
Solution Approach 1:
The patent segments the time-series data into multiple subsequences based on different candidate seasonality patterns (e.g., hourly, daily, weekly patterns). Each subsequence is then independently clustered to evaluate how well it fits the candidate pattern. This segmentation allows the system to break down the complex problem of seasonality detection into manageable parts, reducing overall computational complexity while maintaining detection accuracy.
Solution Approach 2:
The patent applies partial action by using silhouette scores to evaluate only the most promising candidate seasonality patterns rather than exhaustively analyzing all possible patterns. The system computes silhouette scores for a limited set of candidate patterns and selects those above a threshold, avoiding the computational burden of evaluating every possible seasonality pattern while still achieving reliable anomaly detection.
2Adaptability or versatility
If all raw time-series data is stored for later analysis, then data availability and analysis flexibility are improved, but storage requirements and data processing time increase
Solution Approach 1:
The patent extracts only the essential characteristics needed for anomaly detection by computing silhouette scores and identifying seasonality patterns from the raw data. Rather than storing and processing all raw data points, the system extracts key temporal patterns and seasonal characteristics, maintaining analysis flexibility while significantly reducing the volume of data that needs to be retained and processed.
3Productivity
If traditional anomaly detection methods are used without seasonality consideration, then processing speed is improved, but false positive and false negative rates increase
Solution Approach 1:
The patent performs preliminary action by pre-computing silhouette scores for candidate seasonality patterns and identifying the dominant seasonal patterns before conducting anomaly detection. This preliminary analysis of seasonality characteristics allows the system to establish baseline expectations for normal seasonal behavior, enabling faster and more accurate anomaly detection without requiring complex real-time computations during the actual detection phase.
Data Source
AI summary
Computerized methodologies are disclosed that are directed to detecting a seasonality pattern that corresponds to a time-series data set. Operations of one methodology includes for each candidate seasonality pattern of a set of candidate seasonality patterns, partitioning the time-series data set into a set of subsequences according to a date-time pattern of a selected candidate seasonality pattern, clustering data points of each of the set of subsequences into two or more clusters, and determining a silhouette score for the selected candidate seasonality pattern that represents a measure of a clustering quality of the selected candidate seasonality pattern. The seasonality pattern from the set of candidate seasonality patterns is then detected by selecting the candidate selecting having a highest silhouette score of the candidate seasonality patterns. An additional operation may include obtaining the time-series data set as a result of execution of a search query received via a graphical user interface.