Analisis Pengaruh Inisialisasi Bobot Nguyen Widrow dan Learning Rate pada Algoritma Learning Vector Quantization

Abstract

The artificial neural netwrork is a field of science that is growing at the present time. Many research using artificial neural networks as a research object. Artificial neural networks are often used in daily life to perform face recognition, classification, decision making, data compression, up to the field of robotics. One of the algorithms used to classify the class of data is Learning Vector Quantization (LVQ). Learning process on LVQ normally use representative vector weights initialization to obtain final weights to be used as the basis for recognizing a class that included the testing process. The purpose of this study was to determine the accuracy of LVQ algorithm using representative vector weights initialization, Nguyen Widrow weight initialization, and the combination of representative vector weights initialization with Nguyen Widrow weight initialization are applied to the two datasets are datasets balance scale and banknote authentication. Results from this research indicate that the use of Nguyen Widrow weights initialization on testing dataset balance scale produces the highest level of accuracy that is equal to 88.71% and showed the same degree of accuracy in all tests for banknote authentication dataset that is equal to 92.23%.