Abstarct:   Penetration rate, which is defined as the speed required to break the drilled rock by the drill bit, is a critical process for optimizing oil well drilling and the total cost of drilling. In this research, three artificial intelligence models of artificial neural network, support vector regression and random forest were used to estimate the penetration rate from drilling data. The data used to build the models is baxsed on real field data points obtained from 9 wells in southern Iran. The data of well 1, well 2, well 3, well 4, well 5 and well 6 were used for training and testing the models and the data was divided into 70% training and 30% testing. To evaluate the infiltration rate estimation in real time, unseen data sets of well 7, well 8 and well 9 were used for validation. After pre-processing and removing outlier data, six characteristics of flow rate, weight on the drill bit, static pipe pressure, depth, torque and rotation of the drill string were used as input to estimate the penetration rate. The sensitivity analysis showed that the artificial neural network with 1 laxyer, 29 neurons, tan-sigmoid as transfer function and trainlm as training function, support vector regression with lambda mextaparameters in the range of 1-10 to 5-10 and epsilon from 5-10 up to 1 and kernel functions and Gaussuia kernel functions and random forest with 300 decision trees and Bootstrap hyperparameter as the optimal hyperparameter and bagging algorithm are the best models for estimating penetration rate. Penetration rate by training and testing three models of artificial neural network, support vector regression and random forest respectively with average absolute error percentage of 5.01%, 5.79%, 9.7% and correlation coefficient 0.93, 0.92, 0.90 was estimated in the data of three validation wells (7, 8 and 9). Therefore, in this research, the artificial neural network with the average percentage of absolute error is lower, the correlation coefficient is higher, and the root mean square error is lower than the support vector regression and random forest models showed a better performance in estimating the penetration rate.