Bagging in the Lack of Reproducibility of Neural Network Results
Valery Zenkov · 2024
by changing the parameter of the pseudo-random optimization process of the neural network, we generate several variants of training the neural network when solving the classification problem. From them, we obtain bagging - an average solution less susceptible to overfitting. We give an example of performing such bagging and compare it with the classification method based on the Anderson discriminant function and the identically related posterior probability of the class.