The Impact of Data Splitting Strategy on Drilling Rate Prediction in the Rumaila Oil Field

Supervised machine learning is one of the important tools that has helped solve many problems facing humanity, especially problems that cannot be solved by humans. Building a successful and high-accuracy model depends on several things, such as the collected data, choosing the appropriate model, the method of data splitting to be used in training and evaluating the model, and choosing the appropriate hyperparameters. Data splitting is one of the most important things to do to obtain a high-accuracy model and to avoid overfitting, which produces a model with high training accuracy but fails in testing and prediction. This paper investigates the impact of different data splitting strategies such as hold-out with different testing sizes, K-Fold, and shuffle split on the effectiveness of a supervised machine learning model for prediction drilling rate in Rumaila oil field in southern Iraq and selecting the optimal data splitting strategy. The highest testing accuracy obtained was 0.827 when the shuffle split strategy was used.