Comparative Analysis of ARIMA and LSTM Models for Seasonal Times-Series Forecasting

This paper compares Auto-Regressive Integrated Moving Average (ARIMA) and Long Short-Term Memory (LSTM) models for forecasting time series of day-ahead electricity market price data. ARIMA can capture temporal dependencies, while LSTM neural networks detect non-linear short-term temporal dependencies. The authors' comparison of forecast prices obtained by ARIMA and LSTM with actual prices shows which of the methods is more suitable for forecasting seasonal time series data with shifting patterns in energy market prices. According to the error measurement results of the obtained results with mean squared prediction error (MSPE), the Seasonal ARIMA (SARIMA) method shows a slightly better accuracy of 6.37 %, while the LSTM has an MSPE of 6.91 %. Research shows that SARIMA effectively applies seasonal patterns in time series such as electricity price data.