Rainfall Prediction in Central Java Based on Comperatif Analysis of LSTM and BiLSTM

Predicting rainfall accurately is essential to many facets of daily life. Often, rainfall exhibits complex and nonlinear patterns that are difficult for conventional statistical models to fully represent. Due to their effective capacity to describe temporal relationships, Long Short-Term Memory (LSTM) and Bidirectional LSTM (BiLSTM) neural networks have recently drawn attention as possible alternatives. This study uses historical data from Central Java, Indonesia, to assess the accuracy of LSTM and BiLSTM models in predicting seasonal rainfall. The evaluation measures include prediction accuracy under various scenarios, Mean Absolute Error (MAE), and Root Mean Square Error (RMSE). The results indicate that, while both the LSTM and BiLSTM models perform competitively, the BiLSTM model shows somewhat better accuracy metrics under various climatic circumstances and parameter configurations. This work advances rainfall forecasting techniques by demonstrating how BiLSTM networks can improve predictive skills.