Persuasion is difficult. In the joint classifier algorithm (similar to boosting), the method is similar to yours, especially the random subspace method. However, the joint algorithm is only effective for weak classifiers, and it has even been proved that it must be over-adaptive for strong linear classifiers.

When paying attention to the persuasiveness of related literature, it is always said that the joint algorithm may be useful for weak classifiers. The ordinary support vector machine itself is absolutely stable in both classification and regression, so if you do what you say, it is almost unconvincing, which is equivalent to finding data for the algorithm instead of making an algorithm based on the data.