Adaptive Thresholding for Multi-Domain Service Assurance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data monitoring solutions rely on predefined thresholds that fail to adjust to real-time trends, leading to inadequate network performance and inefficient resource allocation due to manual adjustments and poor understanding of network trends.
Innovation Solution
Implementing adaptive thresholding techniques that utilize machine learning models to dynamically adjust thresholds based on data trends, allowing for real-time monitoring and intelligent resource allocation by clustering and processing data points to identify anomalous data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If predefined thresholds are used for data monitoring, then the monitoring system is simple to implement, but the thresholds fail to adjust to real-time trends and cannot account for new normal network states
Solution Approach 1:
The patent implements dynamic threshold adjustment by continuously learning from incoming data streams. Machine learning models update thresholds in real-time based on observed patterns and trends, allowing the system to adapt to changing network conditions without manual intervention. This transforms static predefined thresholds into dynamic, self-adjusting parameters that evolve with the monitored system.
Solution Approach 2:
The monitoring system performs self-service by automatically detecting anomalies and adjusting thresholds without requiring manual configuration or expert intervention. The machine learning models autonomously learn from data patterns, identify deviations from normal behavior, and adapt thresholds accordingly, enabling the system to serve itself and reduce dependency on human operators.
2Reliability
If manual adjustments are made to monitoring thresholds, then the system can respond to changes, but the adjustments reduce monitoring efficacy and allow network performance to degrade
Solution Approach 1:
The system implements continuous feedback loops where monitored data is fed back into machine learning models that adjust thresholds based on observed performance patterns. This closed-loop feedback mechanism enables automatic detection of performance degradation and real-time threshold adaptation, eliminating the delays inherent in manual adjustment processes and maintaining high monitoring reliability.
Solution Approach 2:
The machine learning models perform preliminary action by proactively learning normal network behavior patterns and predicting deviations before they become significant issues. The system prepares adaptive thresholds in advance based on learned patterns, enabling it to quickly respond to anomalies without waiting for manual detection or reaction to performance degradation.
3Measurement precision
If complex machine learning models are implemented for adaptive thresholding, then anomaly detection improves, but the computational resources and system complexity increase
Solution Approach 1:
The patent applies partial action by implementing machine learning models at selective points in the monitoring architecture where they provide maximum value. Rather than applying complex models throughout the entire system, the approach uses adaptive thresholding selectively for critical metrics and anomalies, balancing detection precision with computational overhead by applying intelligence only where most needed.
Data Source
AI summary
Techniques for adaptive thresholding are provided. A first data point in a data stream is received, and a first plurality of data points from the data stream is identified, where the first plurality of data points corresponds to a timestamp associated with the first data point. At least a first cluster is generated for the first plurality of data points, and a predicted value for the first data point is generated based at least in part on data points in the first cluster. A deviation is computed between the predicted value for the first data point and an actual value for the first data point. Upon determining that the deviation exceeds a first predefined threshold, the first data point is labeled as anomalous, and reallocation of computing resources is facilitated based on labeling the first data point as anomalous.