Drilling formation perception by supervised learning: Model evaluation and parameter analysis

Abstract

Formation and lithology identification based on well logging curves reflecting geophysical response characteristics is fundamental for drilling planning and reservoir recovery. For the purpose of providing efficient, accurate, and comprehensive insights for drilling operation decisions, the present research evaluates three typical supervised learning algorithms based on machine learning, e.g. Adaboost, decision tree and support vector machine (SVM). By comparing the prediction results from three typical classification algorithms based on performance metrics such as accuracy, precision, recall, F1 score, Adaboost and decision tree are found to present more accurate prediction with relatively higher accuracy, precision, recall and F1 score. The prediction accuracy is positively related to the training data set proportion for all three approaches. Decision tree approach spends less computation time while still provides favorable prediction scores. The accuracy of prediction gradually increases as the number of logging features increases. Accuracy for most parameter combinations beyond four logging parameters can be up to 90%. The neutron porosity and the spontaneous potential are considered as the most influential parameters affecting the prediction accuracy.