Classification Model Training via Synthetic Minority Over-sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing classification models trained on unbalanced datasets tend to favor classes with a large proportion, leading to prediction errors and poor generalization due to data feature loss or over-fitting, as conventional under-sampling and over-sampling methods often result in data feature loss, noise increase, and marginalization of data distribution.
Innovation Solution
A method that determines sampling rates for classes in a dataset, generates new samples based on Euclidean distances to balance class distributions, and trains the classification model using these new samples, thereby improving prediction accuracy by dynamically adjusting the data distribution feature information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If under-sampling is used to balance class proportions in unbalanced data, then the class distribution becomes more balanced, but data feature losses occur and prediction model accuracy deteriorates
Solution Approach 1:
The patent applies preliminary action by generating new minority class samples before model training using SMOTE (Synthetic Minority Over-sampling Technique). This synthetic over-sampling creates intermediate samples based on Euclidean distances between existing minority samples and their nearest neighbors, thereby balancing class distributions while preserving data features and avoiding the information loss associated with under-sampling.
2Quantity of substance
If simple data duplication is used for over-sampling to balance class proportions, then the class distribution becomes more balanced, but data feature offsets and noise increase occur
Solution Approach 1:
The patent applies parameter changes by transforming the sampling rate parameter dynamically. Instead of simple duplication, it uses SMOTE to generate synthetic samples with adjusted feature values based on Euclidean distances and nearest neighbor relationships. This changes the parameter space of the minority class samples, creating diverse synthetic data that balances class proportions without causing feature offsets or noise increase.
3Quantity of substance
If over-sampling with simple data duplication is used to balance class proportions, then the data distribution becomes more balanced, but over-fitting occurs and generalization ability deteriorates
Solution Approach 1:
The patent applies preliminary action by performing synthetic over-sampling using SMOTE before model training. This preliminary step generates diverse minority class samples through interpolation in feature space, balancing class distributions while maintaining data variability. The generated synthetic samples prevent over-fitting by introducing controlled diversity, thereby improving the model's generalization ability to unseen data.
Data Source
AI summary
Provided are a classification model training method, system, electronic device, and storage medium. The method includes: determining sampling rates of first-class samples and second-class samples in a data set, and setting the samples with a sampling rate less than a preset value as target samples (S101); determining data distribution feature information of the target samples based on Euclidean distances between all the samples in the data set (S102); wherein the data distribution feature information is information describing the number of same-class samples in nearest neighbor samples, and the nearest neighbor samples are two samples at a Euclidean distance less than a preset distance; generating new samples corresponding to the target samples based on the data distribution feature information (S103); and training the classification model using the first-class samples, the second-class samples and the new samples (S104).