Comparative Analysis of the Accuracy of Backpropagation and Learning Vector Quantisation for Pattern Recognition of Hijaiyah Letters

Abstract

The Artificial Neural Network (ANN) is a branch of science in the field of artificial intelligence and is created from adapting the workings of the human brain. Backpropagation (BP) and Learning Vector Quantisation (LVQ) are two of many methods used to recognise patterns. Both are supervised training methods with different approaches. BP uses an error value to recognise patterns or images, while LVQ uses distance values as an indicator classification of patterns in class. This study conducted a simulation of pattern recognition to compare the accuracy of BP and LVQ in terms of recognising Hijaiyah letter patterns (Arabic characters) to see how the number of epochs and the value of learning rate during the training process affects the accuracy. Simulations carried out 28 targets or Hijaiyah letter with parameter epoch 25, 100, 500, 1000, 3000 and 5000 as well as the learning rate of 0.001, 0.05, 0.01, 0.1, 0.25 and 0.05. Overall, the resulting accuracy on Backpropagation method is better compared to LVQ. It is influenced by the value of learning rate and the number of epochs. Accuracy of the Backpropagation training is less influenced by the value of learning rate, but it becomes more accurate if the number of epochs used for training a lot in accordance with the number of samples used. While LVQ, contrary to Backpropagation, the accuracy of the results of training are affected by the value of learning rate and is less affected by the number of epoch.