Integration of extreme gradient boosting feature selection approach with machine learning models: application of weather relative humidity prediction

Relative humidity (RH) is one of the important processes in the hydrology cycle which is highly stochastic. Accurate RH prediction can be highly beneficial for several water resources engineering practices. In this study, extreme gradient boosting (XGBoost) approach “as a selective input parameter” was coupled with support vector regression, random forest (RF), and multivariate adaptive regression spline (MARS) models for simulating the RH process. Meteorological data at two stations (Kut and Mosul), located in Iraq region, were selected as a case study. Numeric and graphic indicators were used for model’s evaluation. In general, all models revealed good prediction performance. In addition, research finding approved the importance of all the meteorological data for the RH simulation. Further, the integration of the XGBoost approach managed to abstract the essential parameters for the RH simulation at both stations and attained good predictability with less input parameters. At Kut station, RF model attained the best prediction results with minimum root mean square error (RMSE = 4.92) and mean absolute error (MAE = 3.89) using maximum air temperature and evaporation parameters. Whereas MARS model reported the best prediction results at Mosul station using all the utilized climate parameters with minimum (RMSE = 3.80 and MAE = 2.86). Overall, the research results evidenced the capability of the proposed coupled machine learning models for modeling the RH at different coordinates within a semi-arid environment.