Predicting rate of penetration (ROP) with machine learning models: A physics-validated comparative case study of Gradient Boosting Regressor (GBR), Random Forest (RF), and K-Nearest Neighbours (KNN)

A successful drilling operation plays a vital role in the profitability of exploration and production of hydrocarbon reserves. However, these operations can come with high drilling rig rental costs. Reliable predictions of Rate of Penetration (ROP) can avoid unnecessary drilling equipment and rig rental expenses. In this study, Gradient Boosting Regressor, Random Forest, and K-Nearest Neighbours were developed to predict rate of penetration based on drilling operational data. The machine learning models were benchmarked with established empirical models, such as the Bingham and Bourgoyne and Young models, which are typically used to predict the rate of penetration. All machine learning models achieved, coefficient of determination (R2) > 0.93 and Average Absolute Percentage Error (AAPE) < 3% on the test dataset. Random Forest under Scenario 1 was identified as the most physically consistent model, attaining strong predictive performance with R2 = 0.9494 & AAPE = 1.96%, satisfying both the Weight on Bit (WOB) flounder point, and rotary speed optimum criteria used within the industry. These results support the practical use of machine learning-based ROP prediction as a reliable tool for real-time optimization of drilling parameters, offering a pathway to reduce non-productive time and rig costs while preserving the physical interpretability that drilling engineers require for field decision making.