Abstarct: Abstract One of the challenges of the drilling industry considering that it is a high-cost and high-risk industry is to improve the speed of drilling by observing safety, economic, technological limitations, equipment, required materials, and naturally, land, which can play a very important role. As the drilling of wells becomes more complex (horizontal, extended, high temperature and pressure, etc.) , oil field researchers conduct more in-depth studies of the factors affecting the penetration rate of the drill bit. The aim of this study is to predict drilling penetration rates and evaluate their impact on different machine learning models using real well data. The South Pars gas field is one of the fields in the south of Iran, which includes two Kangan and Vedalan reservoirs in the Zagros sedimentary area. The Kangan and Dalan formations are related to the Dahram group, which are among the most important gas reservoirs in the Persian Gulf in that area. baxsed on lithology and porosity, the reservoir zone of South Pars field is divided into two parts K-1 and K-2. To achieve this goal, data from 4 rings of directional drilling wells in South Pars gas field have been used as input data. For the data set used in this study, there were more than 4000 data points, so The ratio of 60% to 40% was used for training and test sets. This study showed that the weight on the drill bit is one of the most important factors in predicting the drilling penetration rate. But there are problems to accurately measure this factor in complex wells. Because the values of the weight on the drill, which are measured on the surface of the ground and in the depth of the well, do not match. Therefore, the completed well data were used to identify and extract relevant factors and data. After entering the software space and normalizing between zero and one, they are pre-processed according to the codes of each algorithm. In the discussion of optimizing the value of the drilling rate to maximize the ratio of 12 factors affecting the data, it was evaluated. Then, using a Python code, the amount of weight on the downhole drill affected by the angle of the well was predicted using surface measurements, and the determination coefficient of 0.95 was obtained, which is a favorable result. In this study, after selecting and preparing the data, several machine learning methods were used to find the best model for predicting the penetration rate of the drill bit. These methods included random forest regression, K-nearest neighbors, artificial neural networks, and long short-term memory. The results showed that the long-short-term memory model was the most accurate model. Then, the effect of various surface factors and their combination on the performance of the model was investigated.