Rainfall Prediction Using Long Short-Term Memory Method (Case Study: Jambi City)

Abstract

Rainfall prediction plays a crucial role in various fields, including meteorology, agriculture, and disaster mitigation. However, the accuracy of such predictions is often affected by missing values in historical rainfall data. To address this issue, this study employs three interpolation methods—linear, quadratic, and quadratic spline—to fill in the missing values. These values are then used for rainfall prediction in Jambi City using the LSTM model. The rainfall data used in this study were obtained from the BMKG Online Data portal, covering the period from 2016 to 2024. The preprocessing stages included data cleaning, applying interpolation methods to address missing values, normalizing the data, and splitting the dataset into training and testing sets. The LSTM model was developed with various hyperparameter configurations to achieve optimal performance. The model’s performance was evaluated using the Root Mean Square Error (RMSE) metric to assess the accuracy of the predictions. The experimental results show that linear interpolation yields the best performance, with RMSE Train and Test values of 13.0661 and 13.1388, respectively. The model, configured with 50 neurons, 75 epochs, a batch size of 32, and using the RMSprop optimizer, demonstrated improved prediction accuracy compared to other interpolation methods. This research confirms the importance of proper data preprocessing in improving the accuracy of time series prediction models. With optimal handling of missing values, LSTM-based prediction models can produce more accurate rainfall forecasts, benefiting various sectors that depend on meteorological data. Finally, to facilitate forecasting for non-programmer users, a web-based Graphical User Interface (GUI) was developed using the Streamlit library.