Comparison of LSTM and ARIMA Algorithms in Predicting Water Levels in Jakarta

This study evaluates the efficacy of Long Short-Term Memory (LSTM) and Autoregressive Integrated Moving Average (ARIMA) algorithms in forecasting water levels (TMA) in Jakarta. The dataset encompasses water level measurements from multiple places in Jakarta from January to June 2024, consisting of 55,920 observations divided into 80% training data and 20% testing data using random sampling. Data preparation entails the use of normalization via MinMaxScaler for LSTM, but ARIMA requires stationarity testing and parameter selection guided by the Akaike Information Criterion (AIC). The investigation indicates that ARIMA marginally surpasses LSTM regarding Mean Squared Error (MSE) and Root Mean Squared Error (RMSE), yielding values of 372.32 and 19.30, respectively. Simultaneously, LSTM has a reduced Mean Absolute Error (MAE) of 13.82 in contrast to ARIMA’s 14.69. The LSTM’s Mean Squared Error (MSE) is 384.30, and the Root Mean Squared Error (RMSE) is 19.60. In conclusion, ARIMA is more adept at identifying data patterns related to MSE and RMSE, while LSTM is superior in minimizing average absolute error. The selection of the optimal model is contingent upon the prioritized evaluation metric and the dataset’s complexity.