A Triangular Six-Line Oversampling Method for Unbalanced Datasets

An improved triangle oversampling technique is introduced to address the challenge of data imbalance in industrial classification tasks. In this method, the new sample is synthesized precisely along the median line of the triangle, ensuring that it aligns with the mathematical expectation of the triangle's center. Furthermore, the variance of this newly synthesized sample, under the action of the regulator, closely approximates the variance of the original sample, effectively simulating the density distribution of raw data. The samples produced by this method exhibit a distribution similar to the original samples, and the efficacy of this proposed method has been validated through numerical experiments.