Data Classification for Neural Network Training

Abstract

This study proposes models that solve the pattern recognition problem on the example of x-ray images of bone fractures. These models are based on neural network programming. The training efficiency for these models exceeds the teacher network performance by almost 5 times with insignificant network training quality reduction. Using the proposed technique, the neural network was trained in several ways, including the conventional method. Additionally, the network's binary classification was changed. The deviations of the obtained network values from the predicted responses are minimal, which has formed the base for the introduced concept of the model's indecisiveness and uncertainty. Furthermore, various experiments with the network were made to test these models, which lead to the idea of using data classification for neural network training.