Analisis Tingkat Akurasi Variasi Algoritma Min-Max Backpropagation sebagai Pre-Processing Data Time Series

Abstract

Forecasting data does not have to be static, there are also data with high fluctuations with up and down trends. Therefore, data normalization techniques are very important before training and testing data. This paper aims to test eight types of Min-Max backpropagation algorithms with several types of data, namely static data, seational data, monotonically fluctuating data up and down. A backpropagation network architecture with three hidden layers is used to test these data. The test results show that the 6th Min-Max algorithm has a high level of accuracy. Furthermore, the results of the 6th Min-Max modification found that changes in the multiplier variable can reduce the MSE value in the training process to a maximum value of 35.25% and in the testing process to 27.39%. The results of this study can be used as a reference in the future in performing the data nornalization process before the forecasting process is carried out.

Keywords

• min-max algorithm, data normalization techniques, time series data, backpropagation algorithm