Anomaly Detection in Time Series Data via Forecasting Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional web analytics methods face challenges in efficiently detecting and ranking anomalies in large volumes of multidimensional web traffic data, making it difficult for analysts to identify important events and trends within the data.
Innovation Solution
A server system generates forecasting models for time series data to identify anomalies by determining if values fall outside the error variance of these models, and reports these anomalies to client applications, allowing for efficient detection and visualization of significant events.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional web analytics methods are used to detect anomalies in large volumes of multidimensional web traffic data, then the analysis can be performed using simple visualization techniques, but it becomes difficult to efficiently identify important events and trends due to the vast amount of data and number of configurations
Solution Approach 1:
The patent replaces manual visual analysis of multidimensional web traffic data with automated statistical forecasting models. Multiple forecasting models (e.g., ARIMA, exponential smoothing) automatically analyze time series data, compute error variances, and identify anomalies without requiring analysts to manually examine hundreds of visualized configurations, thereby dramatically improving detection efficiency while managing complexity through algorithmic processing
Solution Approach 2:
The patent creates multiple forecasting models as copies to analyze the same time series data from different statistical perspectives. Each model generates its own error variance estimates, and anomalies are identified when values deviate beyond thresholds across multiple model copies, providing robust anomaly detection without requiring a single complex analysis system
2Reliability
If multiple forecasting models are generated for each attribute to improve anomaly detection accuracy, then the reliability of anomaly identification increases, but the computational complexity and processing time increase
Solution Approach 1:
The patent applies partial action by generating multiple forecasting models only for attributes where anomalies are expected to be meaningful, rather than exhaustively analyzing all possible attribute combinations. The system selectively applies forecasting models to time series data based on data availability and analytical needs, reducing unnecessary computational complexity while maintaining reliable anomaly detection for critical metrics
Solution Approach 2:
The patent segments the analysis process by dividing it into distinct phases: data collection, forecasting model generation, anomaly detection, and result reporting. Each phase handles specific tasks independently, allowing the system to manage complexity through modular processing while maintaining high reliability through comprehensive multi-model analysis in the forecasting phase
Data Source
AI summary
A server system stores time series data for a data source. The time series data comprises a plurality of time-value pairs, each pair including a value associated with an attribute of the data source and a time. For a particular attribute, the server system generates a plurality of forecasting models for characterizing the time-value pairs, each model including an estimated attribute value and an associated error-variance. For a time-value pair, the server system determines a plurality of differences between the value of the time-value pair and respective estimated attribute values of the plurality of forecasting models and tags the time-value pair as an anomaly if the differences for at least a first subset of the forecasting models are greater than the corresponding error variances. In response to a request from a client application, the server system returns at least a subset of the time-value pairs tagged as anomalies.