Graph Autoencoder for Business Process Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autoencoder-based approaches for anomaly detection in business processes fail to consider structural information and treat event logs as simple flat data, lacking the ability to capture key relationships among activity occurrences.
Innovation Solution
The use of a Graph Autoencoder (GAE) approach that converts business process logs into a graphical data structure, applying an unsupervised machine learning method with a graph neural network (GNN) and edge-conditioned convolution to generate optimized graph encodings for anomaly detection, enabling the computation of anomaly scores based on process-aware metrics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current autoencoder-based approaches are used for anomaly detection, then the implementation is simple, but the detection accuracy is low due to ignoring structural information
Solution Approach 1:
The patent transforms the flat event log data into a graph structure by adding a spatial dimension that represents structural relationships between activities. This graph structure captures the hierarchical and relational information that is lost in traditional flat representations, thereby improving anomaly detection accuracy without excessively increasing model complexity.
Solution Approach 2:
The patent introduces graph neural networks as an intermediary component between the input data and the autoencoder. This intermediary transforms the raw event logs into graph representations that preserve structural information, enabling the subsequent autoencoder to achieve better detection accuracy while managing complexity through modular architecture.
2Loss of information
If Graph Autoencoder with GNN is applied, then structural information is captured effectively, but the training convergence becomes more difficult
Solution Approach 1:
The patent performs preliminary construction of the graph structure from event logs before feeding data into the GNN model. By pre-processing the data into graph format with properly defined nodes, edges, and attributes, the model receives well-structured input that facilitates faster and more stable training convergence while effectively capturing structural information.
Solution Approach 2:
The patent optimizes training convergence by adjusting key parameters including learning rate, batch size, and graph construction parameters. These parameter changes help balance the complexity of GNN training with the need to preserve structural information, enabling effective training of the Graph Autoencoder model.
Data Source
AI summary
A method, computer system, and a computer program product for anomaly detection is provided. The present invention may include converting business process logs into a graphical data structure. The present invention may include generating an optimized graph encoding for anomaly detection using an unsupervised machine learning model. The present invention may include computing an anomaly score for each activity of the business process log using a process aware metric based on feature representation. The present invention may include labeling each of the one or more data points with a high anomaly score.