The Effect of Batch Size on the Performance MobileNet-V3 Model for Breast Cancer Classification of Ultrasound and Mammography Images

Many hyperparameters must be adjusted for a Convolutional Neural Network (CNN) to classify pictures accurately. The batch size or the total number of images used to train the forward and backward routes, is one of the most significant hyperparameters. The purpose of this work is to examine how batch size affects Convolutional Neural Networks (CNN) ability to classify images, particularly medical photos that show breast cancer. In this experiment, the MobileNet-V3 model is employed to train the network more quickly. The testing on ultrasound images highest level of accuracy was 98.05%, loss of 0.0496 on training data and a 84.00% accuracy, a loss 0.4772 on validation data with batch size 256. Testing on mammography data the best accuracy, with a value train accuracy of 94.97% and a loss of 0.1265, and validation data accuracy of 90.34% and a loss of 0.3177 with batch size 32