A Comparative Study of ARIMA and LSTM in Forecasting Time Series Data

Forecasting time series data is an important subject in economics and business where the Autoregressive Integrated Moving Average (ARIMA) has been extensively used despite its weaknesses, from requiring a minimum number of data points to the assumed linearity of data. With recent advancement, the Long Short-Term Memory (LSTM) shows potential to address such weaknesses. This research is aimed to identify a more suitable model in handling irregular data. Performance metrics used are the model accuracy measured with RMSE and the model run-time performance measured with the Python Timeit library. This research concluded that LSTM is more accurate than ARIMA (RMSE of ARIMA 0.144887 to LSTM 0.051828) in a shorter dataset of 36 data points and this result is reverted in longer dataset of 228 data points (RMSE of ARIMA 0.006949 to LSTM 0.036025). In terms of run-time performance, ARIMA is significantly faster than LSTM while the LSTM modelling time is increasing proportionally to the number of the training data points.