Cluster-Based Data Augmentation for Imbalanced Machine Learning Datasets
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In machine learning, especially with small or imbalanced datasets, data augmentation is necessary to improve prediction accuracy, but existing methods can be time-consuming and may not effectively enhance the quality and variety of data, potentially leading to overfitting and resource inefficiency.
Innovation Solution
A computer-implemented method for data augmentation that clusters input data into groups based on similarity, determines clusters that require augmentation, and applies specific augmentation methods to improve prediction accuracy, thereby optimizing the augmentation process and reducing computational resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If data augmentation is performed on the entire dataset, then the size and quality of the dataset is improved, but the time consumption and computational resources increase significantly
Solution Approach 1:
The patent divides the dataset into multiple clusters based on feature similarity. Instead of applying data augmentation uniformly to the entire dataset, the method selectively applies augmentation only to specific clusters that meet certain criteria (e.g., small cluster size, high variance, or low representation). This segmentation approach reduces the overall computational burden while maintaining the benefits of data augmentation where most needed.
Solution Approach 2:
The patent applies different data augmentation strategies to different clusters based on their specific characteristics. Each cluster is evaluated independently, and augmentation is applied locally only where necessary. This local quality approach ensures that computational resources are focused on the most beneficial areas of the dataset rather than uniformly processing all data points.
2Quantity of substance
If data augmentation is performed on the entire dataset, then the quality and variety of data is improved, but the computational resources and efficiency deteriorate
Solution Approach 1:
By segmenting the dataset into clusters and selectively applying augmentation only to certain clusters, the method maintains data variety in the most critical areas while preserving computational efficiency overall. The segmentation allows the system to focus resources on generating diverse data where it matters most.
Solution Approach 2:
The patent applies data augmentation partially rather than excessively to the entire dataset. By identifying and augmenting only the clusters that benefit most from this process, the method achieves sufficient data variety without the diminishing returns and resource waste that would result from applying augmentation uniformly across all data.
3Stability of the object's composition
If data augmentation is applied uniformly to all clusters, then the dataset becomes more balanced, but the risk of overfitting increases due to redundant data
Solution Approach 1:
The patent applies data augmentation locally to specific clusters rather than uniformly across all clusters. This allows the method to balance the dataset in areas where it is most needed while avoiding the creation of redundant data in clusters that already have sufficient representation. The local quality approach prevents overfitting by ensuring that augmentation is applied with precision rather than blanket coverage.
4Quantity of substance
If traditional data augmentation methods are used, then the dataset size increases, but the processing time and resource usage increase proportionally
Solution Approach 1:
The patent segments the dataset into clusters and processes only the relevant clusters for data augmentation. This segmentation reduces the total number of data points that require augmentation processing, thereby increasing processing speed while still achieving the desired increase in overall dataset size through targeted expansion of specific clusters.
Data Source
AI summary
A computer-implemented method for data augmentation is provided according an embodiment of the present disclosure. In the method, a first feature vector for input data may be obtained based on a first model. The input data may be clustered to a plurality of clusters. For each of the clusters, a second feature vector may be obtained based on the first model. Then, a similarity between the first feature vector and the second feature vector may be estimated for each of the clusters. At least one cluster of the plurality of clusters for which the similarity is lower than a threshold may be determined. Moreover, data augmentation may be performed to the at least one cluster.