Weather Forecasting in Denpasar City Using Stacked Long Short-Term Memory Algorithm (LSTM)

Abstract

Forecasting the weather is essential to sustaining everyday human activities, particularly in industries like tourism, agriculture, and transportation. The effects of extreme weather events can be lessened by timely and accurate weather forecasts. This research suggests a Denpasar City weather forecasts are made utilizing a deep learning technique and the Stacked Long Short-Term Memory architecture. The four main parameters of the model— temperature, humidity, wind speed, and pressure—were trained using historical weather data spanning 1990 to 2020. A sliding window method was used to organize the dataset into time-series sequences after it had been preprocessed using normalization techniques. The Adam optimizer was used to train the model over 50 epochs with a batch size of 64. Four regression measures were used for evaluation: Coefficient of Determination (R2), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Squared Error (RMSE). The Stacked LSTM model's average MAE of 1.08, MAPE of 10.22%, RMSE of 1.93, and R2 of 0.86 demonstrate how well it captures temporal patterns and generates precise forecasts, according to the experimental data. These results show how the Stacked LSTM approach can be used to support decision-making in weather-sensitive domains and create automated weather forecasting systems.