Weather Prediction in Jakarta: An Analysis of Climate Data and Regional Influences using LSTM and GRU

This research investigates meteorological predictions in Jakarta using Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models, focusing on precipitation forecasting. Meteorological data, encompassing wind speed, temperature, humidity, and precipitation, were sourced from the Kemayoran Meteorological Station in Jakarta. After data pre-processing to ensure suitability for machine learning training, the study contrasted LSTM and GRU models’ performances using Root Mean Square Error (RMSE) and R-squared (R2) as evaluation metrics. Findings showed the GRU model marginally outperformed during testing, though the difference wasn’t statistically significant. The LSTM model exhibited greater consistency, suggesting its potential for more dependable weather predictions given the right dataset. The research underscores the necessity of considering regional factors, climatological data, and data time intervals for accurate forecasting. It offers insights into machine learning’s potential in meteorological predictions. Specifically, at a learning rate of 0.0002, GRU had an RMSE of 18.034, slightly better than LSTM’s 18.075. However, at 0.0003, LSTM outperformed with an RMSE of 17.972 compared to GRU’s 17.988, highlighting the nuanced performances of these models.