PREDIKSI TINGKAT CURAH HUJAN KOTA MATARAM MENGGUNAKAN LONG SHORT-TERM MEMORY

Abstract

Rainfall is an important weather parameter that significantly influences the agricultural sector and regional planning. The City of Mataram, as the center of social and economic activities in West Nusa Tenggara Province, requires an accurate rainfall prediction method. This study aims to predict daily rainfall in Mataram City using a multivariate long short-term memory (LSTM) approach. The data used consist of daily observations from BMKG, with input variables including average temperature, average humidity, average wind speed, and average air pressure. The dataset is structured as a time series using a sliding window approach with a 7-day lookback period and is divided into 80% training data and 20% testing data. The LSTM model is constructed with two LSTM layers containing 64 and 32 units, respectively, complemented by a 0,2 drop out layer and a Dense layer as the output for prediction. Evaluation using MAE and RMSE indicates that a configuration of 100 epochs and a batch size of 16 provides the best performance, achieving MAE of 4,020 mm and RMSE of 7,915 mm on the testing data, demonstrating the model’s capability to predict daily rainfall in a stable manner.