Outlier Detection Models Using Balanced Synthetic Event Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Artificial intelligence models trained on imbalanced data, particularly those with a low percentage of outlier samples, suffer from high false positives due to ignored or obscured features of outlier events, leading to poor quality in outlier detection.
Innovation Solution
Implement an unsupervised machine learning model trained using balanced data sets by oversampling outlier events, either with actual or synthetic data, to improve accuracy and efficiency in outlier detection, utilizing both a majority and minority expert model for enhanced classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If a model is trained on imbalanced data with many more regular samples than outlier samples, then the model can be trained with sufficient data volume, but the outlier detection quality deteriorates with high false positives
Solution Approach 1:
The patent applies oversampling to create synthetic outlier samples, intentionally generating more outlier data points than naturally exist. This excessive action on the minority class balances the training data distribution, allowing the model to learn outlier patterns without being dominated by regular samples, thereby reducing false positives while maintaining sufficient overall data volume
Solution Approach 2:
The patent changes the data distribution parameter by transforming imbalanced training data into balanced training data through oversampling. This parameter change in class distribution enables the model to achieve better outlier detection accuracy without sacrificing the benefit of having sufficient total data for training
2Ease of manufacture
If sample outlier events are excluded or insufficiently represented in training data, then the training process becomes simpler, but the model loses the ability to accurately detect outlier features
Solution Approach 1:
The patent creates copies of existing outlier samples through synthetic data generation techniques. By copying and transforming outlier event patterns into multiple synthetic samples, the model gains sufficient representation of outlier features without requiring complex manual annotation or extensive collection of rare outlier events, thus maintaining training simplicity while improving reliability
Solution Approach 2:
The patent performs preliminary action by pre-generating synthetic outlier samples before model training. This advance preparation ensures that outlier features are adequately represented in the training data from the outset, eliminating the need for complex handling during training and ensuring the model learns accurate outlier patterns without compromising training simplicity
Data Source
AI summary
Methods and systems are described herein for outlier detection. The system may apply a single-tier including a minority expert model only or a two-tier machine learning model, including a majority expert model and a minority expert model. The system may generate the minority expert model by training an unsupervised machine learning model on oversampled training data, including synthetic sample outlier events. In some embodiments, the minority expert model may provide a binary result indicating an event belongs to an outlier category or not. In some embodiments, the minority expert model may include multiple component models providing a multi-class result indicating whether an event belongs to a sub-category of the outlier category. In application, the system may perform the outlier detection on events that are sequence-based or not.