Early IT Failure Detection Using Multimodal Metric Correlation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing IT ticketing systems lack the ability to predict upcoming issues and failures based on multiple tickets and often generate non-actionable, temporal, and unnecessary tickets, hindering effective incident management.
Innovation Solution
Implementing a machine learning-based failure detection model that correlates metric patterns in time series data to predict upcoming IT failures and identifies non-actionable tickets by utilizing rule-based decision making and artificial intelligence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional ticketing systems are used to track IT incidents, then basic incident tracking is achieved, but the system cannot predict upcoming failures and generates many non-actionable tickets
Solution Approach 1:
The system performs preliminary analysis of historical ticket data and system metrics to predict potential failures before they occur. The failure prediction model analyzes patterns in past incidents and current system state to generate early warnings, enabling proactive remediation before actual failures impact service.
Solution Approach 2:
The system creates a virtual representation of the IT system state by copying and analyzing historical ticket data and metrics patterns. This virtual model allows the system to simulate and predict failure scenarios without affecting the actual system, enabling safe prediction and analysis.
2Productivity
If all IT tickets are processed equally, then comprehensive incident coverage is achieved, but non-actionable tickets consume valuable IT agent time
Solution Approach 1:
The system extracts and separates non-actionable tickets from the overall ticket stream using classification models. These tickets are identified based on patterns matching known non-actionable scenarios (such as transient issues already resolved or duplicate reports), allowing them to be filtered out and excluded from standard ticket processing workflows.
Solution Approach 2:
The system implements feedback loops where ticket resolution outcomes and agent actions are fed back into the classification model. This continuous learning process improves the accuracy of identifying non-actionable tickets over time, as the model learns from historical data which ticket patterns led to non-actionable outcomes.
3Measurement precision
If historical ticket data is analyzed in detail to improve prediction accuracy, then failure detection precision is improved, but data processing time and complexity increase
Solution Approach 1:
The system segments the analysis process into distinct components: data collection, preprocessing, pattern recognition, and prediction. Historical ticket data is divided into relevant features (ticket outcomes, resolution times, error patterns) and fed to specialized processing modules. This segmentation allows complex analysis to be performed in manageable stages, reducing overall system complexity.
Data Source
AI summary
Embodiments relate to early detection of information technology (IT) failures in a computing system. A technique is executed by one or more processors and includes receiving multiple IT records including past and recent historical data, extracting first and second time series sections from the past and recent historical data, respectively, training a failure detection model to correlate metric patterns in the first time series sections with at least one of other metric patterns and previous IT failures and, in response to the training, using the failure detection model to predict at least one of upcoming metric patterns and upcoming IT failures from metric patterns in the second time series sections.