Air relative humidity (RH) is a key component of the hydrological cycle, and its accurate prediction is vital for all fields related to global climate change. In this study, the performances of six intelligent models, including outlier-robust extreme learning machine (ORELM), weighted regularized extreme learning machine (WRELM), regularized extreme learning machine (RELM), standard extreme learning machine (ELM), bootstrap aggregating (Bagging), and adaptive boosting (AdaBoost), was evaluated in daily RH prediction. Two weather stations (Tiaret and M’sila) located in Algeria were selected as a case study. The models were developed using data collected for the period ranging from 2000 to 2013. The performance of the models were evaluated using Nash–Sutcliffe efficiency (NSE), correlation coefficient (R), mean absolute error (MAE), and root-mean-square error (RMSE). Based on the statistical criteria, the accuracy ranks of the six models were: ORELM, WRELM, RELM, ELM, Bagging, and AdaBoost at both stations. The best results are obtained using the ORELM model with R, NSE, RMSE, and MAE of approximately ≈ 0.935, ≈ 0.874, ≈ 7.943, and ≈ 6.135 at Tiaret station, and ≈ 0.928, ≈ 0.860, ≈ 7.955, and ≈ 6.132 at M’sila station, respectively. Overall, ORELM represents a novel method that can effectively model the RH with high predictive accuracy.