<![CDATA[Severe weather conditions such as fog and heavy precipitation pose significant threats to aviation safety. Accurate prediction of aircraft visibility is therefore essential to support operational decision-making and reduce the likelihood of accidents. This study aims to compare and evaluate the performance of two bidirectional deep learning models, BiLSTM and BiGRU, in predicting aircraft visibility using historical meteorological data from BMKG Juanda Sidoarjo. The novelty of this research lies in applying and comparing bidirectional recurrent architectures for visibility prediction, an approach rarely explored in aviation meteorology, to assess their capability in capturing temporal dependencies within time-series visibility patterns. Both models were trained using hyperparameter tuning, with the best configuration obtained from a 24-hour input window, batch size of 32, 64 neurons, a dropout rate of 0.1, and 100–200 epochs. The dataset was divided into training and testing sets (80:20), and model performance was evaluated using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE) to assess both predictive accuracy and computational efficiency. The results indicate that while BiLSTM achieved slightly higher accuracy, BiGRU demonstrated superior overall efficiency, obtaining competitive error metrics (MSE = 1.50 × 10⁶, RMSE = 1,223.5, MAPE = 19.35%) compared to BiLSTM (MSE = 1.58 × 10⁶, RMSE = 1,258.1, MAPE = 19.50%). BiGRU’s advantage lies in its simpler structure and faster computation, which reduce training complexity without sacrificing forecast accuracy. Overall, this research contributes to the development of efficient bidirectional time-series models for aviation meteorology, offering a practical framework for real-time visibility forecasting in computationally limited environments. The balance between accuracy, speed, and model simplicity makes BiGRU a more scalable and applicable choice for enhancing flight safety operations.]]>
