Generating combined feature embedding for minority class upsampling in training machine learning models with imbalanced samples

This invention describes a way to create new, artificial training examples for machine learning models, especially when there aren't enough examples of a particular type. It works by taking two existing training examples, blending their data and their correct answers together using a random ratio, to form a new, combined example. This newly created, blended example is then used to teach the machine learning model, helping it learn better from scarce data. The claims specifically detail how to blend the data and the correct answers from two existing samples to generate a new training sample.