The Impact of Varying Batch Sizes on the Deep Neural Network Algorithms Performance

Batch size is one of the determining factors in the quantity of sample data used in one iterative model training in deep learning. One of the key factors influencing the effectiveness of hyperparameters and model training results. Group size is very important in selection because the right group size will influence the effectiveness and performance of the training. Choosing a batch size that is too small or large can affect accuracy and performance. Deep Neural Network (DNN) algorithm is a form of artificial neural network consisting of several deep layers. There are several batch sizes used in data processing. Optimal performance for batch size is obtained when using a batch size of 64, as it yields the highest average AUC value of 0.815. Furthermore, the optimal configuration was achieved with a batch size of 64 and 6 hidden layers, utilizing a combination of Leaky-ReLV and Sigmoid activation functions, resulting in an AUC value of 0.83.