Stacking Ensemble Modeling of Bidirectional LSTM and Bidirectional GRU for Air Temperature Prediction in Ngawi

Abstract

Artificial Neural Networks (ANN) have rapidly developed and are used in forecasting, classification, and regression by mimicking how the human brain processes data. Recurrent Neural Networks (RNN), such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), are effective in processing sequential data and handling long-term dependencies. Bidirectional LSTM (BiLSTM) and Bidirectional GRU (BiGRU) process data in both directions to enhance accuracy. This study evaluates the implementation of the Stacking Ensemble method using BiLSTM and BiGRU as base learners and Random Forest as the meta learner to predict air temperature in Ngawi Regency. Air temperature prediction is crucial as it affects agriculture, health, and energy sectors. The data used comprises 2282 records from January 1, 2028, to March 31, 2024, processed using Google Colab. The results show that the Stacking BiLSTM-BiGRU model with Random Forest provides the best performance with a Mean Squared Error of 0.0005, Root Mean Squared Error of 0.0233, Mean Absolute Error of 0.0179, and R-squared of 0.9832, outperforming other individual models. This study confirms that the Stacking Ensemble method with BiLSTM and BiGRU significantly improves air temperature prediction accuracy.