Air Temperature Prediction System Using Long Short-Term Memory Algorithm

Air temperature is a highly essential parameter in weather forecasting methods and a critical variable for predicting future weather patterns. An accurate temperature prediction system can assist individuals and organizations in preparing for activities heavily influenced by weather conditions. Therefore, developing a precise temperature prediction model requires a reliable and effective algorithm. In this study, the Long Short-Term Memory (LSTM) algorithm, a type of artificial neural network (Recurrent Neural Network - RNN), is implemented with time series data decomposition for variable input processing. LSTM is specifically designed to handle sequential data or time series data, such as weather data. Additionally, LSTM-GRU and LSTM-Conv1D models are utilized. The dataset used in this research comprises air temperature data provided by the Meteorology, Climatology, and Geophysics Agency (BMKG) in the DKI Jakarta region. Model evaluation is conducted using criteria for the smallest Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). Experiments show that the prediction system based on LSTM-GRU achieves the lowest MAE and RMSE values compared to LSTM and LSTM-Conv1D, across 10, 20, and 30-step predictions. It can be concluded that the LSTM-GRU algorithm provides the most accurate predictions compared to the LSTM and LSTM-Conv1D models for sequential temperature data, given sufficient data and a properly configured model. This is also graphically demonstrated by prediction results closely aligning with the actual data.