Prediction of Iran's monthly average temperature with time series methods and regionalizing area using ANN, LSTM, and GRU artificial neural networks

Molavi-Arabshahi, Mahboubeh

doi:10.22034/ijcoe.2026.602936.1280

Prediction of Iran's monthly average temperature with time series methods and regionalizing area using ANN, LSTM, and GRU artificial neural networks

Document Type : Original Research Article

Author

Mahboubeh Molavi-Arabshahi

School of Mathematics and Computer Science, Iran University of Science and Technology, P.O.Box 1684613114, Narmak, Tehran, Iran.

10.22034/ijcoe.2026.602936.1280

Abstract

This paper investigates the accuracy of five artificial neural networks, namely ANN, LSTM, and GRU, in predicting monthly temperature data. The study employs the monthly temperature data set of 31 meteorological stations in Iran as its time series data set, which is partitioned into five regions with distinct climates to ensure accurate temperature predictions. Three metrics, including MSE, MAE, and R2Score, are used to measure the prediction error of each neural network. The results indicate that the Stacked LSTM neural network outperforms other artificial neural networks in terms of predicting monthly temperature data, making it the most appropriate method for this purpose. Notably, the ANN neural network exhibits the lowest accuracy in predicting monthly temperature data among the examined neural networks. This study highlights the effectiveness of Stacked LSTM neural networks in time series analysis and prediction and provides valuable insights for the application of artificial neural networks in this domain.

Keywords

Neural network

Monthly temperature

ANN

LSTM

GRU

Subjects