Weather Prediction Using CNN-LSTM-Based AI for Weather Pattern Analysis in Banten Province

Abstract

Weather variability in Banten Province poses challenges across various sectors, including community activities, agriculture, and disaster preparedness, necessitating accurate weather prediction methods. This study proposes a hybrid Convolutional Neural Network–Long Short-Term Memory (CNN-LSTM) model to predict air temperature and rainfall based on historical weather time-series data. The dataset was obtained from the Open-Meteo API and BMKG for the observation period from January 2019 to May 2024. Input variables include air temperature, rainfall, wind speed, sea surface temperature anomaly, and the El Niño–Southern Oscillation (ENSO) index. The data preprocessing stages involve data cleaning, normalization using Robust Scaler, and the construction of data sequences using the sliding window method prior to the model training process. Model performance was evaluated using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) and compared against baseline models. The experimental results demonstrate that the CNN-LSTM model achieves an MAE of 0.60°C and an RMSE of 0.73°C for air temperature prediction and an MAE of 6.15 mm and an RMSE of 8.31 mm for rainfall prediction. The prediction outcomes were subsequently integrated into a web-based dashboard to facilitate information visualization. Initial validation against BMKG observation data in South Tangerang showed a relatively low temperature deviation during the testing period. These findings confirm that the proposed approach has adequate potential to support short-term weather prediction systems in Banten Province.