SVDD Outlier Detection Using Window Distance Calculation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Support vector data description (SVDD) methods face significant computational challenges when dealing with large training datasets, leading to high computing times due to the need to solve quadratic programming problems, which hampers efficient outlier detection and classification.
Innovation Solution
A system is implemented that determines support vector data descriptions for outlier identification by computing a set of support vectors, Lagrange constants, center and control limit values, and using these to identify outliers through a window distance calculation, thereby reducing computational complexity and improving efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If SVDD is used for outlier detection in large datasets, then detection accuracy is improved, but computing time increases significantly
Solution Approach 1:
The patent divides the SVDD computation into two distinct phases: a training phase where the SVDD model is built from historical data, and a monitoring phase where the pre-computed model is applied to detect outliers in real-time. This segmentation separates the computationally intensive model construction from the rapid inference process, enabling accurate outlier detection without the full computational burden of retraining on every new data point.
Solution Approach 2:
The patent performs preliminary computation of the SVDD model parameters (support vectors, Lagrange constants, center, and radius) during the training phase using historical data. These pre-computed parameters are then stored and reused during the monitoring phase, eliminating the need to perform the complex quadratic programming problem solving again for each new observation, thus significantly reducing real-time computing time.
2Measurement precision
If the number of observations in training dataset is increased, then outlier detection accuracy is improved, but computing time increases
Solution Approach 1:
The patent performs the computationally intensive SVDD computation once during the training phase using historical data, and stores the resulting model parameters for reuse. This preliminary action allows the system to leverage large training datasets for high accuracy without incurring the full computational cost repeatedly during monitoring operations.
Solution Approach 2:
The patent creates a simplified representation of the training data in the form of pre-computed SVDD parameters (support vectors, Lagrange constants, center, radius) that capture the essential characteristics of the training dataset. This copied representation allows the system to maintain the benefits of large training datasets while reducing the computational complexity of subsequent operations.
3Measurement precision
If SVDD model parameters are computed for every observation, then detection accuracy is maintained, but computational complexity increases
Solution Approach 1:
The patent segments the computational tasks into training and monitoring phases, computing the full SVDD model only during training and using simplified distance calculations during monitoring. This segmentation maintains detection accuracy while significantly reducing the complexity of operations performed on each new observation.
Solution Approach 2:
The patent performs the complex SVDD parameter computation as a preliminary action during the training phase, before any monitoring occurs. The pre-computed parameters are then used for simple distance calculations during monitoring, maintaining accuracy while reducing per-observation computational complexity.
Data Source
AI summary
A computing device identifies outliers. Support vectors, Lagrange constants, a center threshold value, an upper control limit value, and a lower control limit value are received that define a normal operating condition of a system. The center threshold value, the upper control limit value, and the lower control limit value are computed from the vectors and the Lagrange constants. A first plurality of observation vectors is received for a predefined window length. A window threshold value and a window center vector are computed. A window distance value is computed between the window center vector and the support vectors. Based on comparisons between the computed values and the received values, the first plurality of observation vectors is identified as an outlier relative to the normal operating condition of the system. When the first plurality of observation vectors are identified as the outlier, an alert is output.