Selection of the most accurate training algorithm in the backpropagation network based on the accuracy of data pattern matching Available to Purchase

Backpropagation is a supervised learning algorithm that is widely used to solve problems. There are 12 types of training algorithms contained in the Backpropagation network. These algorithms need to be tested to get the most accurate training algorithm in the accuracy of data pattern matching. In this study, the twelve algorithms were tested using two neuron models, namely 15-20-1 and 15-25-1. In this test, the main network parameters used are target error=0.001 with 12 variations of learning rate, namely 0.01, 0.05, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, and 1. Based on the results of the ANOVA test with alpha (α)=5%, the results show that the Levenberg-Marquardt (LM) algorithm is the most accurate algorithm for both neuron models with an average difference between the network output data and the target (delta) of 0.00577 and 0.00515, respectively. This delta value is achieved at lr=0.2 for the neuron 15-20-1 model and lr=0.8 for the 15-25-1 neuron model. Thus, the LM algorithm can be considered in choosing a training algorithm for application development in the field of artificial neural networks.