Analisis Data Time Series Menggunakan LSTM (Long Short Term Memory) Dan ARIMA (Autocorrelation Integrated Moving Average) Dalam Bahasa Python.

Abstract

Abstract — This study aims to predict time series data by using two methods, the first method commonly used is statistics Autocorrelation Integrated Moving Average (ARIMA model) and the second method which is relatively new, namely machine learning Long Short Term Memory (LSTM). Before the data is processed by both methods, data cleaning and data optimization are carried out. Data optimization is a transformation process to eliminate elements of trends and variations from data. The transformation consists of 7 results of a combination from Log processes, Moving Average (MA), Exponential Weigh Moving Average (EWMA), and Differencing (Diff). The seven processes are each used in the ARIMA and LSTM processes. So that 14 predictions will be obtained (7 from the ARIMA process and 7 from the LSTM process). From the 14 prediction results obtained the smallest RMSE value for ARIMA is 2% and the smallest RMSE value for LSTM is 1%. The results of this study using 7 combinations of transformation processes, can increase the level of accuracy of predictions from ARIMA and LSTM. Where the accuracy of LSTM learning machines by using Telkom's stock data has higher accuracy than ARIMA.