Predicting Temperatures in an Extreme Climatic Environment Using Hybrid Neural Networks: Evaluating Noise Robustness

Ali W. Alattabi (Department of Civil Engineering, Wasit University, Wasit 52001)
Salah L. Zubaidi (1) Department of Civil Engineering, Wasit University, Wasit 52001, Iraq. 2) College of Engineering, University of Warith Al-Anbiyaa, Karbala 56001)
Hussein Al-Bugharbee (Department of Mechanical Engineering, Wasit University, Wasit 52001)
Hussein Mohammed Ridha (4) Advanced Lightning, Power and Energy Research (ALPER), Department of Electrical and Electronics Engineering, Faculty of Engineering, Universiti Putra Malaysia, Serdang 43400, Malaysia. 5) Department of Computer Engineering, University of Al-Mustan)
Mawada Abdellatif (Department of Civil Engineering and Built Environment, Faculty of Engineering Technology, Liverpool John Moores University, Byrom Street, Liverpool L3 3AF)
Hassimi Abu Hasan (7) Department of Chemical and Process Engineering, Faculty of Engineering and Built Environment, Universiti Kebangsaan Malaysia, UKM, Bangi, Selangor 43600, Malaysia. 8) Research Center for Sustainable Process Technology (CESPRO), Faculty of Engineer)

Civil Engineering Journal
Vol. 12 No. 5 (2026): May

Predicting maximum temperatures is crucial across many fields and industries, including medicine, agriculture, energy, and climate research. Researchers have not treated the prediction of maximum temperatures under severe artificial data disturbance in much detail. So, it has not yet been understood. This research aims to integrate an artificial neural network (ANN) with the Guaranteed Convergence Arithmetic Operation Algorithm (GCAOA) to forecast monthly maximum temperatures while ensuring robustness to noise. Univariate data from Al-Hai City over 12 years were employed to build and assess the model. The performance of GCAOA was examined and compared with that of the two hybrid ANNs, the random forest, and the XGBoost models. Across various input scenarios, the results reveal that these three hybrid models achieved very good forecast performance compared with random forests and XGBoost. The GCAOA-ANN (swarm size of 20 and lag2) achieves the best forecast performance among the hybrid algorithms across different statistical fitness measures with a coefficient of determination, Nash-Sutcliffe coefficient, and root mean squared error of 0.972, 0.969, and 1.7354°C, respectively. The performance of the hybrid ANN models was further investigated under noise, and the results showed the superiority of the GCAOA-ANN model.