UndersamplingIt is a method to alleviate class imbalance by discarding samples, which can be understood as under-sampling the categories with a large number of samples in the training set.
The purpose of undersampling in the data is to solve the problem of uneven categories in the sampling process. In addition to undersampling and oversampling, it is also feasible to combine the two techniques into a hybrid strategy, such as SMOTE and Tomek link, SMOTE and edited nearest neighbor ENN.
Other approaches to learning on imbalanced datasets, such as weighing training instances, introduce different misclassification and bootstrapping costs for positive and negative examples.
Related terms: oversampling, class imbalance
References
【1】https://www.cnblogs.com/Determined22/p/5772538.html
【2】https://en.wikipedia.org/wiki/Oversampling_and_undersampling_in_data_analysis