Local energy loss is among the essential parameters of culvert design, in which uncertainty and nonlinearity is controversial. In the present study, seven models were developed with the aid of the experimental data of slope-tapered culverts, and the efficiency of gene expression programing and Gaussian process regression as a kernel-based approach was assessed in predicting the entrance loss coefficient of a slope-tapered culvert. Also, one-at-a-time (OAT) sensitivity analysis was performed to determine the impact of input parameters. The results of both GEP and GPR methods with the performance criteria of R = 0.847, DC = 0.777, RMSE = 0.2 and R = 0.76, DC = 0.718, RMSE = 0.25 showed that the model with input parameters of Froude number (Fr), ratio of headwater to culvert diameter (Hw/D) and ratio of reducer length to barrel length (Lr/L) is the superior model. Although the accuracy of GEP method was slightly higher than GPR, obtained results proved the capability of the applied methods (i.e., high correlation coefficient (R) and coefficient of determination R2 (DC) and low RMSE). Furthermore, OAT sensitivity analysis revealed that Froude number has the most impact on local loss coefficient and could cause a significant increment in model efficiency.