WEATHER PREDICTION USING A COMPARISON OF LSTM AND GRU METHODS

Abstract

Accurate weather condition prediction is essential to support various community activities, particularly in tropical regions where weather patterns are relatively complex. This study was conducted to evaluate and compare the capabilities of Long Short-Term Memory and Gated Recurrent Unit in classifying weather conditions into three categories: sunny, cloudy, and rainy. The research data consisted of monthly climatological data from January 2021 to December 2025 obtained from the South Sumatra Climatology Station. The data were processed using linear interpolation to generate daily data comprising 1,796 observations. The research process included data cleaning, normalization, class encoding, sequence formation using a sliding window, training and testing data splitting, and model evaluation based on accuracy, precision, recall, and F1-score. The results showed that the Gated Recurrent Unit achieved an accuracy of 88%, while the Long Short-Term Memory achieved 77%. Therefore, the Gated Recurrent Unit demonstrated superior performance in weather condition classification.