Implementasi Sliding Window Algotihm pada Prediksi Kurs berbasis Neural Network

Abstract

Time series is sequential data based on time sequence. Time series data can be used for prediction topics, one of the prediction topics that is always interesting to study is exchange rate prediction. In the case of exchange rate prediction, an appropriate data preprocessing stage is required. The success of this preprocessing stage will have a major effect on the resulting RMSE value. There is an important technique in determining the best RMSE value, especially in time series data, one of which is the windowing technique. The windowing technique is the stage of transforming time series data into cross sectional. Window size has an important role in time series data. However, there is no standard in window size. The Window size experiment starts with a small value and then increases to a larger value until it reaches a certain point with the best RMSE. In this research, an experiment will be conducted on windows size on exchange rate data based on a neural network. The purpose of this research is to optimize the RMSE of a data mining model based on windows parameters. The implementation of sliding windows is carried out in the scenarios of window sizes 4, 6, and 28. Based on the experiments conducted, the best RMSE is on windows size 6 = 0.014 +/- 0.000. With a combination of neural network parameters in the form of training cycles = 1000, learning rate = 0.1 and momentum = 0.1.