TY - JOUR AU - Sari, Yuslena AU - Arifin, Yudi Firmanul AU - Novitasari, Novitasari AU - Faisal, Mohammad Reza PY - 2022/10/01 Y2 - 2024/03/29 TI - Deep Learning Approach Using the GRU-LSTM Hybrid Model for Air Temperature Prediction on Daily Basis JF - International Journal of Intelligent Systems and Applications in Engineering JA - Int J Intell Syst Appl Eng VL - 10 IS - 3 SE - Research Article DO - UR - https://ijisae.org/index.php/IJISAE/article/view/2184 SP - 430–436 AB - <p>Air temperature has a rapid change movement every day. Temperature prediction is very important as a proper reference base for decision making and good planning for stakeholders. However, in the time series daily temperature prediction, the right number of input combinations has not been found for high accuracy. To overcome this, we propose a deep learning approach using the Hybrid gated recurrent units (GRU) - long short-term memory (LSTM) model. These two deep learning models are very suitable for time series predictions. It has 2 (two) main advantages, namely: A variety of input scenarios is used to find the most reliable performance. (1) the model eliminates the time series decomposition process by embedding a time layer to achieve efficient predictions, and (2) the model achieves a stronger high-level temporal to produce reliable performance. Performance measurement uses root mean squared error (RMSE), mean absolute error (MAE), and R-Squared (R2). Best RMSE on 15-day input, which is 0.07499. The best result of MAE is with a value of 0.0578 at the input of 15 days. The performance results obtained both RMSE and MAE, the smallest of the 15 experimental scenarios is at the input of 15 days. The results of R2 are in line with the results of RMSE and MAE, namely the input in 15 days produces the best R2 close to 1, which is 0.9937.</p> ER -